hash stringlengths 32 32 | doc_id stringlengths 5 12 | section stringlengths 5 1.47k | content stringlengths 0 6.67M |
|---|---|---|---|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 5.1 Key Issue #1: Authorization for AIMLE Service Security for AIML members
| |
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 5.1.1 Key issue details
|
3GPP TS 23.482 [3] introduces support for AIMLE services, enabling AI/ML operations through interactions between the AIMLE client and AIMLE server(s) over the AIML-UU reference point, and between the VAL servers and AIMLE servers over AIML-S respectively. These services involve distributed AI/ML operations across multiple participants, necessitating robust security mechanisms to ensure that only authorized members participate in the AIMLE workflows. Further, TR 23.700-83 [4] describes AIMLE based MI model inference and AI Inference exposure services. Given the critical role of authorization in securing these workflows, it is important to assess whether the current security specifications are adequate.
Currently, the authorization aspects outlined in TS 33.434 [2] can be limited to address the security requirements of AIMLE services and related aspects specified in TS 23.482 [3] such as related to a) Federated Learning (FL), b) client related handling (registration, discovery, selection, selection subscription, and participation), c) transfers (task transfer, transfer learning, context transfer) d) ML Model (training capability evaluation, monitoring and control), e) Split operations and AIMLE assistance respectively. Therefore, this key issue aims to study whether enhancements to the authorization mechanisms specified in 3GPP TS 33.434 [2] are necessary to support AIMLE service security. The objective is to ensure trusted AIMLE members participation and usage to prevent unauthorized access of AIMLE operations.
Editor’s Note 1: Rel-20 related security aspects need to consider and align with the conclusions in TR 23.700-83 [4] is FFS.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 5.1.2 Security threats
| Unauthorized AIMLE members (e.g., FL members) participating in AIMLE services may gain access to data exchanged between AIMLE clients and servers.
Lack of robust authorization allows unreliable or unauthorized AIMLE members (e.g., FL members) to degrade the quality, efficiency, or availability of AIMLE operations.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 5.1.3 Potential security requirements
| The 3GPP system shall support authorization mechanisms for AIML members (e.g., FL members) utilising AIMLE services for various AIMLE procedures.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 5.2 Key Issue #2: Secure AIMLE ML Model Access
| |
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 5.2.1 Key Issue details
| TS 23.482 [3] describes AIMLE services which supports ML Model retrieval, ML model training, ML model management (model information storage and discovery) ML model update, and ML model selection aspects. AIMLE Services uses SEAL as the fundamental architecture and the authorization aspects of SEAL Security in TS 33.434 [2] which allows requested service specific authorization which can be limited and necessary controls can be in place for the different ML access and management work flow authorization for the overall AIMLE based ML access security.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 5.2.2 Security threats
| Unauthorized AIMLE client(s)/ VAL server using AIMLE services may gain access to ML model data leading to leakage of model.
Lack of robust authorization allows unauthorized AIMLE client(s) or VAL servers to degrade the quality, efficiency, or availability of AIMLE operations.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 5.2.3 Potential security requirements
| The 3GPP system shall support authorization to secure AIMLE service-based ML Model operations such as retrieval, training, update, selection, and management (i.e., ML model information storage and discovery).
5.X Key Issue #X: <Key Issue Name>
5.X.1 Key Issue details
5.X.2 Security threats
5.X.3 Potential security requirements
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6 Solutions
| Editor’s Note: This clause contains the proposed solutions addressing the identified key issues.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.1 Solution #1: Authorization for AIMLE Services
| |
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.1.1 Introduction
| This solution address KI#1.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.1.2 Solution details
| AIMLE authorization related to AIML Services can reuse the authorization procedure specified in TS 33.434 [2] clause 5.2.2 (SEAL service authorization) and clause B.3.3 (SEAL service authorization) as the baseline where, SIM-S or AIMLE Server (with SIM capabilities) acts as an authorization server and issues access token to the AIMLE service consumer. The AIMLE service producer provides the requested services to the AIMLE service consumers by verifying the authorization of AIMLE service consumer i.e., on validating the access token claims as shown in Figure 6.1.2-1.
Figure 6.1.2-1: AIMLE Service Authorization
Step 1-3. The access token request, access token generation, response can be same as TS 33.434 [2] Clause B.3.7 Obtaining access token and B.3.6 Access token, with the adaptation that scope includes AIMLE service specific information.
Step 4-6. The AIMLE service Request/Response sent is same as each of request/response messages described in TS 23.482 [3] clause 8 related procedures with the following adaptations. i.e., The access token is sent in step 4 and on successful validation of AIMLE service specific information in the access token claims, the AIMLE service request is processed, and the response is provided.
The specific authorization related adaptations to AIMLE Service related procedures include the following:
1. FL member registration:
• AIMLE Service: FLMemberRegistration Request/Response, FLMemberRegistration Update Request/Response, FLMemberRegistrationFetch Request/Response, FLMemberDeregistration Request/Response
• AIMLE Service Consumer: VAL Server, AIMLE Server
• AIMLE Service Producer: ML Repository
• Token Claims including scope: FL member ID/Requestor ID (i.e., AIMLE Service Consumer ID) as Subject, AIMLE service-related information as scope, FL member type (as Server or Client), FL member capabilities, Allowed ML Model ID list, FL member location information, Issuer as Authorization Server ID (i.e., SIM-S ID or AIMLE Server ID).
2. FL related events subscription:
• AIMLE Service: FLEvents Subscribe/Notify
• AIMLE Service Consumer: VAL Server, AIMLE Server
• AIMLE Service Producer: ML Repository
• Token Claims including scope: FL member ID/Requestor ID as Subject, AIMLE service-related information as scope, FL member Type (Server or Client), Allowed FL member ID, Allowed FL related Events ID or name, Allowed ML Model ID list/ML Model Information for FL, Allowed notification target address, issuer as authorization server ID.
3. HFL Training:
Process 1:
• AIMLE Service: MLModelTraining Request/Response
• AIMLE Service Consumer: VAL Server
• AIMLE Service Producer: AIMLE Server
• Token Claims including scope: Requestor ID as Subject, AIMLE service-related information as scope, AIML Model (e.g., Model ID/Type) and Model parameters, Dataset ID(s), Allowed FL members (Allowed List of member client IDs) to use as AI MLE clients for HFL (or) ML model training, Training Type (HFL/VFL/or both), Allowed AI MLE client selection/filtering criteria, Allowed ML Model ID list/ML Model Information for training, ML Model selection filtering criteria, issuer as authorization server ID
Process 2:
• AIMLE Service: HFLTraining Subscribe/Notify
• AIMLE Service Consumer: AIMLE Server
• AIMLE Service Producer: AIMLE Client
4. VFL Training:
Process 1:
• AIMLE Service: MLModelTraining Request/Response
• AIMLE Service Consumer: VAL Server
◦ AIMLE Service Producer: AIMLE Server
• Token Claims including scope: Requestor ID as Subject, AIMLE service-related information as scope, Allowed FL members (Allowed List of member client IDs) to use as AIMLE clients for VFL model training (e.g., per domain), Training Type (HFL/VFL/or both), Allowed AI MLE client selection/filtering criteria, Allowed ML Model ID list/ML Model Information for training, VFL Model selection filtering criteria, issuer as authorization server ID
Process 2:
• AIMLE Service: HFLTraining Subscribe/Notify
• AIMLE Service Consumer: AIMLE Server
• AIMLE Service Producer: AIMLE Client
• Token Claims including scope: Requestor ID as Subject, AIMLE service-related information as scope, Allowed FL members (Allowed List of member client IDs) to use as AI MLE clients for VFL model training (e.g., per domain), Training Type (HFL/VFL/or both), Allowed AI MLE client selection/filtering criteria, Allowed ML Model ID list/ML Model Information for training, VFL Model selection filtering criteria, issuer as authorization server ID
5. FL member grouping:
Process 1:
• AIMLE Service: FLMemberGroupSupport Request/Response
• AIMLE Service Consumer: VAL Server
• AIMLE Service Producer: AIMLE Server
• Token Claims including scope: Requestor ID as Subject, AIMLE service-related information as scope, VAL service ID, AIML Model ID, ADAE Analytics ID, ML Model Profile Information (e.g., ID for which the FL grouping is to be used), ML Task Information/ID (e.g., FL Training task or FT Inference Task), Allowed FL members (Allowed List of member client IDs) to use as AI MLE clients/server for FL, issuer as authorization server ID
Process 2:
• AIMLE Service: FLGroupIndication Request/Response
• AIMLE Service Consumer: AIMLE Server
• AIMLE Service Producer: AIMLE Client
• Token Claims including scope: Requestor ID as Subject, AIMLE service-related information as scope, VAL service ID, AIML Model ID, ADAE Analytics ID, ML Model Profile Information (e.g., ID for which the FL grouping is to be used), ML Task Information/ID (e.g., FL Training task or FT Inference Task), Allowed FL members (Allowed List of member client IDs) to use as AI MLE clients/server for FL, issuer as authorization server ID.
6. AIMLE Client Discovery:
• AIMLE Service: AIMLEClientDiscovery Request/Response
• AIMLE Service Consumer: VAL Server
• AIMLE Service Producer: AIMLE Server
• Token Claims including scope: Requestor ID as Subject, AIMLE service-related information as scope, Allowed maximum number of AIMLE clients, Allowed AIMLE Client discovery criteria such as List of allowed VAL service(IDs), Allowed service permission level usages (premium resource usage/standard resource usage/limited resource usage), Allowed ML model types (decision trees/linear regression/neutral networks/any model type), Allowed AIML operations/services (such as training, model transfer, model inference, model offload, model split), Allowed dataset requirements or handling, Allowed client location/Allowed location information for member client discovery/selection (Anywhere or by coordinates, civic addresses, network areas, or VAL service area ID), Allowed AIMLE Client task capabilities, issuer as authorization server ID.
7. AIMLE Client Registration:
• AIMLE Service: AIMLEClientRegistration Request/Response, Update, Delete
• AIMLE Service Consumer: AIMLE Client
• AIMLE Service Producer: AIMLE Server
• Token Claims including scope: Requestor ID as Subject, AIMLE service-related information as scope, Allowed client profile(s), List of allowed VAL service(IDs) and allowed corresponding permission level(s), Allowed AIML operations/services (such as training, model transfer, model inference, model offload, model split), Allowed client location/Allowed location information for member client selection (Anywhere or by coordinates, civic addresses, network areas, or VAL service area ID), AIMLE Client capabilities, Allowed ML Model ID list/ML Model Information for AIMLE client usage, issuer as authorization server ID.
8. AIMLE Client Selection:
• AIMLE Service: AIMLEClientSelection Request/Response
• AIMLE Service Consumer: VAL Server
• AIMLE Service Producer: AIMLE Server
• Token Claims including scope: Requestor ID as Subject, AIMLE service-related information as scope, Allowed maximum number AIMLE clients, List of allowed VAL service(IDs), Allowed AIMLE Client IDs, Allowed AIMLE client selection criteria i.e., [service permission level usages (premium resource usage/standard resource usage/limited resource usage), Allowed ML model types (decision trees/linear regression/neutral networks/any model type), Allowed AIML operations/services (such as training, model transfer, model inference, model offload, model split), Allowed dataset requirements or handling, Allowed client location/Allowed location information for member client discovery/selection, (Anywhere or by coordinates, civic addresses, network areas, or VAL service area ID), Allowed AIMLE Client task capabilities,], Allowed AIMLE Client Set ID(s), issuer as authorization server ID.
9. AIML Client selection subscription and notification:
• AIMLE Service: AIMLEClientSelection Subscribe/Notify, Update, Unsubscribe
• AIMLE Service Consumer: VAL Server
• AIMLE Service Producer:
• Token Claims including scope: Requestor ID as Subject, AIMLE service-related information as scope, List of allowed VAL service(IDs), Allowed AIMLE client selection criteria/service requirements per VAL service ID i.e., [service permission level usages (premium resource usage/standard resource usage/limited resource usage), Allowed number of AIMLE Clients for selection, Allowed Notification endpoint for the selected AIMLE Client, issuer as authorization server ID.
10. AIMLE Client Participation:
◦ AIMLE Service: AIMLEClientParticipation Request/Response
◦ AIMLE Service Consumer: AIMLE Server
◦ AIMLE Service Producer: AIMLE Client
◦ Token Claims including scope: Requestor ID as Subject, AIMLE service-related information as scope, List of allowed VAL service(IDs), Allowed AIMLE Client Set ID(s), Allowed AIMLE server ID(s), Allowed operation (Add/remove indicator), Allowed AIML model ID(s), Allowed AIML operations/services (such as training, model transfer, model inference, model offload, model split), Allowed AIMLE client selection criteria/service requirements per VAL service ID i.e., [service permission level usages (premium resource usage/standard resource usage/limited resource usage), Allowed dataset requirements or handling, issuer as authorization server ID.
11. AIML Task Transfer:
• Type 1: AIMLE Service: (i) AIMLTaskTransferAssist Request/Response, (ii) AIMLESControlled AIMLTaskTransfer Request/Response
• AIMLE Service Consumer: AIMLE Client
• AIMLE Service Producer: AIMLE Server
• Token Claims including scope: Requestor ID as Subject, AIMLE service-related information as scope, List of allowed VAL service(IDs) and allowed corresponding permission level(s), Allowed AIML Task type or operations/services (such as training, model transfer, model inference, model offload, model split), Allowed ML Model ID list/ML Model Information for AIMLE client usage, Allowed Transfer mode (i.e., AIMLE server assisted/direct AIML task transfer/AIMLE server-controlled), Allowed time window for task transfer, Allowed AI/ML target members for task transfer, issuer as authorization server ID.
• Type 2: AIMLE Service: (i) AIMLTaskTransfer Request/Response, (ii) DirectAIMLTaskTransfer Request/Response
• AIMLE Service Consumer: (i)AIMLE Server (ii)AIMLE Client
• AIMLE Service Producer: AIMLE Client
• Token Claims including scope for (i): Requestor ID as Subject, AIMLE service-related information as scope, Allowed list of Source AI/ML Member ID(s), Allowed list of Target AI/Member ID(s), List of allowed VAL service(IDs) and allowed corresponding permission level(s), Allowed AIML Task type or operations/services (such as training, model transfer, model inference, model offload, model split), Allowed ML Model ID list/ML Model Information for AIMLE client usage, Allowed Transfer mode (i.e., AIMLE server assisted/direct AIML task transfer/AIMLE server-controlled), Allowed time window for task transfer, Allowed AI/ML target members for task transfer, issuer as authorization server ID.
• Token Claims including scope for (ii): Requestor ID as Subject, AIMLE service-related information as scope, List of allowed VAL service(IDs) and allowed corresponding permission level(s), Allowed AIML Task type or operations/services (such as training, model transfer, model inference, model offload, model split), Allowed ML Model ID list/ML Model Information for AIMLE client usage, Allowed Transfer mode (i.e., AIMLE server assisted/direct AIML task transfer/AIMLE server-controlled), Allowed time window for task transfer, Allowed AI/ML target members for task transfer, issuer as authorization server ID.
12. AIMLE Context Transfer:
◦ AIMLE Service: ContextTransfer Request/Response
◦ AIMLE Service Consumer: AIMLE Server (e.g., S-EAS)
◦ AIMLE Service Producer: AIMLE Server (e.g., T-EAS)
◦ Token Claims including scope: Requestor ID as Subject, AIMLE service-related information as scope, Allowed service area information related to the source Edge AIMLE Server ID(s), Allowed list of Target Edge AIMLE Server ID(s) and service area information for context transfer, Allowed list of Target AIMLE Client ID(s) for which context transfer is to done, AIMLE context transfer services (request/response) as scope, List of allowed VAL service(IDs) and allowed corresponding permission level(s), Allowed AIML Task type or operations/services (such as training, model transfer, model inference, model offload, model split), Allowed ML Model ID list/ML Model Information for AIMLE client usage, List of Previous managing AIMLE server ID(s), issuer as authorization server ID.
13. AIML service operations control and management procedure:
Process 1
• AIMLE Service: AIMLEServiceOperationsManagement Request/Response
• AIMLE Service Consumer: VAL Server
• AIMLE Service Producer: AIMLE Server
• Token Claims including scope: Requestor ID as Subject, AIMLE service-related information as scope, Allowed target AIMLE Client identifiers, Allowed target AIMLE Client set identifiers, Allowed AIML service operation identifiers (such as model training id, ml task id etc.), Allowed AIML service operation information, Allowed AIML service operation mode (such as start, stop), Allowed AIML service operation mode configuration, Allowed AIML service operation mode status reporting (such as periodic/event based), List of allowed VAL service(IDs) and allowed corresponding permission level(s), Allowed AIML Task type or operations/services (such as VFL/HFL), issuer as authorization server ID.
Process 2
• AIMLE Service: AIMLEClientServiceOperations Request/Response
• AIMLE Service Consumer: AIMLE Server, AIMLE Client
• AIMLE Service Producer: AIMLE Client
• Token Claims including scope: Requestor ID as Subject, AIMLE service-related information as scope, Allowed target AIMLE Client identifiers, Allowed target AIMLE Client set identifiers, Allowed AIML service operation identifiers (such as model training id, ml task id etc.), Allowed AIML service operation information, Allowed AIML service operation mode (such as start, stop), Allowed AIML service operation mode configuration, Allowed AIML service operation mode status reporting (such as periodic/event based), List of allowed VAL service(IDs) and allowed corresponding permission level(s), Allowed AIML Task type or operations/services (such as VFL/HFL), issuer as authorization server ID.
14. Transfer Learning Enablement
◦ Type 1: AIMLE Service: TLModelSelectionAssistance Request/Response
◦ AIMLE Service Consumer: VAL Server
◦ AIMLE Service Producer: AIMLE Server
◦ Token Claims including scope: Requestor ID as Subject, AIMLE service-related information as scope, List of allowed VAL service(IDs) and allowed corresponding permission level(s), Allowed ML task ID(s), Allowed ADAE analytic ID(s), Allowed ML model profile, Allowed TL criteria, Allowed List of VAL UE ID(s), Allowed ML model requirement information, Allowed list of ML models (ML repository ID and address, ML model rating) for AIMLE service consumer, Allowed Transfer mode (i.e., VAL server triggered /UE triggered), Allowed AI/ML target members for transfer learning, issuer as authorization server ID.
◦ Type 2: AIMLE Service: UE TLModelSelectionAssistance Request/Response
◦ AIMLE Service Consumer: AIMLE Client
◦ AIMLE Service Producer: AIMLE Server
◦ Token Claims including scope: Requestor ID as Subject, AIMLE service-related information as scope, List of allowed VAL service(IDs) and allowed corresponding permission level(s), Allowed ML task ID(s), Allowed ADAE analytic ID(s), Allowed ML model profile, Allowed TL criteria, Allowed List of VAL UE ID(s), Allowed ML model requirement information, Allowed list of ML models (ML repository ID and address, ML model rating) for AIMLE service consumer, Allowed Transfer mode (i.e., VAL server triggered /UE triggered), Allowed AI/ML target members for transfer learning, issuer as authorization server ID.
Editor’s Note: Further details on how the parameters included in the token are used during the authorization verification by the resource server is FFS.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.1.3 Evaluation
| The solution uses the SEAL service authorization procedure as baseline with the following impacts:
To secure the SEAL based AIMLE Services, this solution provides enhancements to the access token claims (such as scope and audience) to indicate AIMLE procedure and information flow specific information to allow related verification at the AIMLE Service producer side before providing any service to AIMLE service consumers.
Editor’s Note: Additional evaluation is FFS.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.2 Solution #2: Authorization of AIMLE clients acting as FL members for access to AIMLE Service Security
| |
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.2.1 Introduction
| This solution proposes the authorization of AIMLE clients in support of federated learning (FL). It ensures that only authorized clients (FL members) are selected, and that secure token-based verification is performed using authorization server. Tokens include only the minimum required claims such as ML model ID / Application Data Analytics Enablement (ADAE) analytics ID and ML model interoperability information to maintain security while ensuring interoperability.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.2.2 Solution details
| |
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.2.2.1 The procedure for AIMLE clients’ authorization
|
Figure 6.2.2.1-1 Authorization’s procedure of AIMLE clients acting as FL members
1. The VAL server sends a FL member grouping support request to the AIMLE server. The request includes the requestor ID, security credentials, and FL grouping criteria (e.g., grouping method, member selection criteria). The initial request is to create the FL member grouping support as described in Step 1 of clause 8.17.2 of TS 23.482 [3]. The security credentials authenticate the VAL server's identity and authorize the grouping request, with validation by the AIMLE server.
2. Upon receiving the request, the AIMLE server validates whether the requestor is authorized to make it.
3. If authorized, the AIMLE server performs an FL member registration fetch with the ML repository based on the FL grouping criteria (see Step 3 of clause 8.17.2 in TS 23.482[3]).
4. The AIMLE server monitors AIMLE clients (FL members) to check whether they meet the selection criteria from step 1 as described in Step 4 of clause 8.13.2.2 of TS 23.482 [3]. AIMLE server interacts with NEF and/or SEAL services (including SEALDD) to set up monitoring. For location-based criteria, it uses SEAL-LMS (3GPP TS 23.434 [5] clauses 9.3.11/9.3.12) or 5GC services (e.g., NEF) to detect UEs entering or present in the target area.
5.a. Using monitoring results, the AIMLE server selects clients that meet the criteria and removes those that do not (e.g., due to location changes).
NOTE 1: The frequency at which monitoring results are provided is left to the implementation.
5.b. Each selected AIMLE client requests an access token from the AIMLE Server. The access token request sent to the AIMLE Server includes the following parameters: ML model ID / ADAE analytics ID and ML model interoperability information.
5.c. Upon receiving the request, the AMILE server issues the generated access token to the client.
6.a. If AIMLE client obtains the access token, the AIMLE client sends a service request message to AIMLE server, requesting the AIMLE server to join FL group. The message contains the ML model ID / ADAE analytics ID and ML model interoperability information, and access token.
6.b. The AIMLE server performs token verification. The AIMLE server obtains the ML model ID / ADAE analytics ID and ML model interoperability information contained in the access token and verifies whether they match the corresponding values in step 5.c.
6.c. In case of successful access token verification, AIMLE server retains the client.
7. The AIMLE server performs the FL member grouping, notifies selected AIMLE clients of their group membership, collects acknowledgements, and returns a FL member grouping support response to the VAL server that includes success/failure status, grouping details, and an optional expiration time for the grouping.
Editor’s Note: Whether and how an AIMLE server can perform the role of authorization server and token validator is FFS.
Editor’s Note: How the solution addresses the authorization aspects of AIMLE client related AIMLE service procedures in TS 23.482 [3] is FFS.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.2.3 Evaluation
| This solution ensures that only authorized AIMLE clients participate as members in FL process. It introduces a token-based authorization process handled by the AIMLE server with support from SEAL.
Editor’s Note: Further evaluation is FFS.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.3 Solution #3: Re-using existing mechanisms
| |
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.3.1 Introduction
| This solution addresses key issue #1 (Authorization for AIMLE Service Security for AIML members) and key issue #2 (Secure AIMLE ML Model Access) by re-using existing mechanisms available in SEAL security architecture.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.3.2 Solution details
| There is no new interface specified for AIMLE services.
Editor’s Note: Further analysis and clarification on interfaces are FFS.
Thus, security for all the interfaces used in the AIMLE has already been addressed including the authorization aspects. Finer granular authorization such as who can be involved in the FL or who can access which ML model can be done locally at the server by using local policy.
Editor’s Note: Clarification on available security for AIMLE interfaces is FFS.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.3.3 Evaluation
| TBD
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.4 Solution #4: Authorization for Secure AIMLE based ML Model Access
| |
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.4.1 Introduction
| This solution address KI#2.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.4.2 Solution details
| AIMLE authorization related to AIML Services can reuse the authorization procedure specified in TS 33.434 [2] clause 5.2.2 (SEAL service authorization) and clause B.3.3 (SEAL service authorization) as the baseline where, SIM-S or AIMLE Server (with SIM capabilities) acts as an authorization server and issues access token to the AIMLE service consumer. The AIMLE service producer provides the requested services to the AIMLE service consumers by verifying the authorization of AIMLE service consumer i.e., on validating the access token claims as shown in Figure 6.1.2-1 and related step description in Clause 6.1.2 (See Solution #1).
The specific authorization related adaptations to AIMLE Service-related procedures include the following:
1. ML Model retrieval:
◦ AIMLE Service: MLModelRetrieval Request/Response, Subscribe/Notify, UpdateSubscription, Unsubscribe
◦ AIMLE Service Consumer: AIMLE Client, VAL Server
◦ AIMLE Service Producer: AIMLE Server
◦ Token Claims including scope: Requestor ID (i.e., AIMLE Service Consumer ID) as Subject, AIMLE service-related information as scope, VAL service ID/information, VAL server ID/VAL UE ID/AIMLE Client ID etc., allowed ML model retrieval filters, and Model Information such as ML Model ID(s)/address or Analytics IDs, audience claim as (AIMLE server /AIMLE Model repository for the AIMLE related services), Issuer as Authorization Server ID (i.e., SIM-S ID or AIMLE Server ID).
2. ML Model Training:
◦ AIMLE Service: MLModelTraining Reqeust/Response
◦ AIMLE Service Consumer: VAL Server
◦ AIMLE Service Producer: AIMLE Server
◦ Token Claims including scope: Requestor ID (i.e., AIMLE Service Consumer ID) as Subject, AIMLE service-related information as scope,Allowed Training Type (VFL or HFL or both VFL and HFL), Allowed List of AIMLE member client IDs, Allowed location information for member client selection, Allowed ML Model ID list/ML Model Information for training, allowed ML model training notification target address, ML Model selection filtering criteria, Issuer as Authorization Server ID (i.e., SIM-S ID or AIMLE Server ID).
3. ML Model Management:
◦ AIMLE Service: (i) ModelInformationStorage Request/Response (ii) ModelInformationDiscovery Request/Response
◦ AIMLE Service Consumer: AIMLE Server, (VAL Server, AIMLE Client (via AIMLE Server))
◦ AIMLE Service Producer: ML Repository
◦ For (i) Token Claims including scope: Requestor ID (i.e., AIMLE Service Consumer ID) as Subject, AIMLE service-related information as scope, Allowed ML information (Model ID, Model Type, or ML model identified by Analytics ID (or) ML model address from where ML model can be downloaded), Allowed AIMLE client ID (or) ML model source identifier (e.g., VAL server ID, VAL client ID, Target ML repository information), Related VAL service ID(s), ML model address, base Model ID, analytics ID, Model size, domain information, Related Vendor ID(s), Issuer as Authorization Server ID (i.e., SIM-S ID or AIMLE Server ID).
◦ For (ii) Token Claims including scope: Requestor ID (i.e., AIMLE Service Consumer ID) as Subject, AIMLE service-related information as scope, VAL service ID, ML model source identifier, Allowed List of AIMLE member client IDs, Allowed location information for member client selection, Allowed ML Model ID list/ML Model Information for discovery, ML model address, base Model ID, analytics ID, allowed ML model training notification target address, ML Model selection filtering criteria, domain information, required accuracy level, Issuer as Authorization Server ID (i.e., SIM-S ID or AIMLE Server ID).
4. ML Model Training Capability Evaluation:
◦ AIMLE Service: MLModelTrainingCapabilityEva Reqeust/Response
◦ AIMLE Service Consumer: AIMLE Server
◦ AIMLE Service Producer: AIMLE Client
◦ Token Claims including scope: Requestor ID (i.e., AIMLE Service Consumer ID) as Subject, AIMLE service-related information as scope, Allowed availability time for supporting FL operations, Allowed test task information, Allowed AI/ML model ID and model parameters, Allowed dataset requirements (such as common feature ID(s), Data domain feature ID(s) list, Data source), etc., List of allowed VAL service(IDs) and allowed corresponding permission level(s), Allowed AIML Task type or operations/services (such as VFL/HFL), Issuer as Authorization Server ID (i.e., SIM-S ID or AIMLE Server ID).
5. ML Model Update:
◦ AIMLE Service: MLModelUpdate Request/Response
◦ AIMLE Service Consumer: VAL server, ADAE server, AIMLE client
◦ AIMLE Service Producer: AIMLE Server
◦ Token Claims including scope: Requestor ID (i.e., AIMLE Service Consumer ID) as Subject, AIMLE service-related information as scope, Allowed ML Model ID, Allowed performance degradation information, Allowed ML model retrieval endpoint (such as URL, URI, IP address), delegated ML model information discovery service via AIMLE server ID(s) list, Issuer as Authorization Server ID (i.e., SIM-S ID or AIMLE Server ID).
6. ML Model performance monitoring:
◦ AIMLE Service: MLModelPerfMonitor Subscriber/Notify
◦ AIMLE Service Consumer: VAL Server
◦ AIMLE Service Producer: AIMLE Server
◦ Token Claims including scope: Requestor ID (i.e., AIMLE Service Consumer ID) as Subject, AIMLE service-related information as scope, Allowed ML Model ID, Allowed Notification endpoint (such as URL, URI, IP address), Allowed AIML operation information (such as ML model training, VFL, HFL, TL etc.), List of VAL service ID, List of AIMLE client ID(s), AIMLE service KPI, Allowed monitoring report configuration, Allowed area of interest, Allowed validity time period, Allowed trigger actions, Issuer as Authorization Server ID (i.e., SIM-S ID or AIMLE Server ID).
7. AIMLE assisted ML Model selection:
◦ AIMLE Service: AssistedMLModelSelection Subscriber/Notify
◦ AIMLE Service Consumer: VAL Server
◦ AIMLE Service Producer: AIMLE Server
◦ Token Claims including scope: Requestor ID (i.e., AIMLE Service Consumer ID) as Subject, AIMLE service-related information as scope, Allowed AIML Profile (such as list of Allowed ML Model ID(s), ML model requirements, Allowed AIMLE Client set ID(s), Allowed AIMLE Client selection criteria, Allowed number of AIMLE clients, Allowed data set ID(s), Allowed Training requirements, Allowed Notification target endpoint (such as URL, URI, IP address), Allowed Notification settings etc.,), Delegated list of AIMLE Server IDs (to perform candidate ML model selection service, ML model information storage service etc., for the AIMLE service consumers), List of VAL service ID, Resource Owner ID as GPSI etc., Audience as AIMLE Server IDs, Issuer as Authorization Server ID (i.e., SIM-S ID or AIMLE Server ID).
Editor’s Note: Further details on how the parameters included in the token are used during the authorization verification by the resource server is FFS.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.4.3 Evaluation
| The solution uses the SEAL service authorization procedure as baseline with the following impacts:
To secure the SEAL based AIMLE Services, this solution provides enhancements to the access token claims (such as scope and audience) to indicate AIMLE procedure and information flow specific information to allow related verification at the AIMLE Service producer side before providing any service to AIMLE service consumers.
Editor’s Note: Additional evaluation is FFS.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.5 Solution #5: FL member authorization for AIMLE services
| |
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.5.1 Introduction
| As specified in TS 23.482[3], the FL members consuming the AIMLE services are AIMLE or VAL server or VAL clients. There are several procedures defined such as ML model retrieval, ML model training, FL member registration, event subscription, AIMLE client registration/discovery/selection/participation so on.
As most of the interaction is between AIMLE client to AIMLE server or VAL server to AIMLE server, it is proposed to re-use the SEAL and VAL service authorization procedure as specified in TS 33.434[2].
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.5.2 Solution details
| For any interaction between AIMLE clients and the AIMLE server, the AIMLE client is provided with the access token by the SIM-S as specified in 5.2 of clause 33.434[2].
For any interaction between VAL server and AIMLE server (e.g., model training), the VAL server is provisioned with an access token by out of band means which is scoped for accessing AIMLE server same as defined for VAL server accessing SEAL key management services in clause 5.3 of TS 33.434[2].
Editor’s Note: Clarification on interaction between VAL server and AIMLE server is FFS.
For any interaction between VAL server or AIMLE server and ML repository (e.g., FL member registration), the VAL server or AIMLE server is provisioned with an access token by out of band means which is scoped for accessing ML repository same as defined for VAL server accessing SEAL key management services in clause 5.3 of TS 33.434[2].
For the procedures like AIMLE client selection/participation or FL member (AIMLE clients) grouping, the AIMLE client is already registered towards ML repository/AIMLE server through SEAL service authorization as specified in 5.2 of clause 33.434[2], no additional authorization procedure is required.
Editor’s Note: Who performs the role of Authorization Server is FFS.
Editor’s Note: Further details on how the solution addresses the overall scope of AIMLE procedures between AIMLE members (FL members) related to KI#1 is FFS.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 6.5.3 Evaluation
| TBD
6.Y Solution #Y: <Solution Name>
6.Y.1 Introduction
Editor’s Note: Each solution should list the key issues being addressed.
6.Y.2 Solution details
6.Y.3 Evaluation
Editor’s Note: Each solution should motivate how the security requirements of the key issues being addressed are fulfilled.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 7 Conclusions
| |
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 7.1 Key Issue #1: Authorization for AIMLE Service Security for AIML members
| Editor’s Note: This clause contains the agreed conclusions for Key Issue #1.
|
d3002a76697e4bc670268c4c89d2da07 | 33.786 | 7.2 Key Issue #2: Secure AIMLE ML Model Access
| Editor’s Note: This clause contains the agreed conclusions for Key Issue #2.
Annex A: Change history
Change history
Date
Meeting
TDoc
CR
Rev
Cat
Subject/Comment
New version
2025-08
SA3#123
S3-252919
AIMLE Service Security TR Skeleton
0.0.0
2025-09
SA3#123
S3-253004
Included Contributions: S3-253003, S3-253005 S3-253006
0.1.0
2025-10
SA3#124
S3-253701
Included Contributions: S3-253134, S3-253697, S3-253698, S3-253699, S3-253700
0.2.0
2025-11
SA3#125
S3-254535
Included Contributions: S3‑254569, S3‑254570, S3‑254571, S3‑254572, S3‑254573, S3‑254574, S3‑254575 , S3‑254727
0.3.0
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 1 Scope
| This document studies potential security and privacy architecture and procedures for 6G mobile networks for improvement of existing services and support of new services, to meet the 6G system requirements and architecture.
One goal of this document is to study how to create lean and streamlined standards for 6G, e.g. by dimensioning an appropriate set of functionalities, minimizing the adoption of multiple options for the same functionality, avoiding excessive configurations, etc.
The document covers the following aspects:
• Security and privacy for overall 6G system architecture
• Security and privacy of 6G RAN architecture.
• Security and privacy of 6G UE to core network interactions.
• Enhancements to Core Network security including endpoint security at transport and application layers, internal and external interfaces as well as end to end roaming security taking roaming intermediary into account.
The document covers possible security enhancements of the procedures from previous generations and new security aspects.
The complete or partial conclusions of this study are used as basis for the normative work.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 2 References
| The following documents contain provisions which, through reference in this text, constitute provisions of the present document.
- References are either specific (identified by date of publication, edition number, version number, etc.) or non‑specific.
- For a specific reference, subsequent revisions do not apply.
- For a non-specific reference, the latest version applies. In the case of a reference to a 3GPP document (including a GSM document), a non-specific reference implicitly refers to the latest version of that document in the same Release as the present document.
[1] 3GPP TR 21.905: "Vocabulary for 3GPP Specifications".
[2] 3GPP TR 38.914: “Study on 6G Scenarios and requirements”.
[3] 3GPP TR 38.760-2: “Study on 6G Radio RAN2 aspects”.
[4] 3GPP TR 23.801-01: “Study on Architecture for 6G System”.
[5] 3GPP TR 33.771: “Study on supporting AEAD algorithms”.
[6] 3GPP TR 23.801-1: "Study on Architecture for 6G System".
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 3 Definitions of terms, symbols and abbreviations
| |
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 3.1 Terms
| For the purposes of the present document, the terms given in TR 21.905 [1] and the following apply. A term defined in the present document takes precedence over the definition of the same term, if any, in TR 21.905 [1].
example: text used to clarify abstract rules by applying them literally.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 3.2 Symbols
| For the purposes of the present document, the following symbols apply:
<symbol> <Explanation>
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 3.3 Abbreviations
| For the purposes of the present document, the abbreviations given in TR 21.905 [1] and the following apply. An abbreviation defined in the present document takes precedence over the definition of the same abbreviation, if any, in TR 21.905 [1].
<ABBREVIATION> <Expansion>
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 4 Security areas and high level security requirements
| |
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 4.1 Security areas
| Editor's Note: This clause further clarifies the scope of the study by listing the security areas that SA3 is working on.
This document includes the following security areas:
1. Security Architecture deals with aspects such as identifying the different security domains and their characteristics, defining the different security functions, etc.
2. RAN security deals with the security aspects of 3GPP access network, e.g., RAN architecture, protocol stack, interfaces, procedures, interaction with UEs.
3. UE to Core Network Security deals with the UE to Core Network communication security. e.g., management of UE and network NAS security contexts, the associated key hierarchy, key derivation and key usage in the 6G System.
4. Core Network Security TBD
5. Subscription Authentication and Authorization deals with different aspects of authentication, authorization and related privacy aspects (i.e. subscriber identifier privacy) for UEs accessing 6G network regardless of access type (i.e., 3GPP access and/or non-3GPP access).
6. Exposure Security deals with security and privacy aspects of 3GPP network exposure.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 4.2 Potential high level security requirements
| Editor's Note: This clause will document high-level requirements that guide the study.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5 Key issues
| |
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.1 Security area #1: Security architecture
| |
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.1.1 Introduction
| This security area addresses the security principles, features, and requirements inherent to the security and trust architecture of 6G systems. This will lay the foundation for all the procedures and the mechanisms necessary to protect the communication and facilitate trust establishment between the UE and the network as well as within/across different domains of the network. The security architecture defined herein provides the foundation for all other security work and is integral to the overall 6G system architecture. The baseline for the work here is to be aligned with the architectural framework described in 3GPP TR 23.801-01 [6].
This includes the following:
- Identifying the different security domains and their characteristics.
- Identifying the security functions, e.g., the security anchors.
- Developing the key hierarchy.
NOTE: Key issues specific to other areas (e.g., aspects that affect the security between the UE and core network) are not intended to be covered in this clause.
Editor’s Note: Privacy aspects in the architecture are FFS.
5.1.2 Security assumptions
Editor's Note: This clause will document security assumptions related to each security area.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.1.3 Key issues
| Editor’s note: This clause will contain the key issues that need to be addressed by SA3 on each security area. The exact contents are FFS.
5.1.3.y Key issue #1.y: <key issue name>
Editor's Note: Key issues within the security area are not in any particular order but they are added incrementally (y = 1, 2, 3…) when new key issue is identified. 'x' refers to the security area.
5.1.3.y.1 Key issue details
5.1.3.y.2 Security threats
5.1.3.y.3 Potential security requirements
5.1.3.y.4 Interim agreements
Editor's note: This clause will include the principles that are agreed as work progresses for the specific KI#x.y. This may be populated directly or e.g. also when a topic in Area #x gets resolved and a principle is agreed. Where there is consensus, interim agreements pertaining to this key issue (e.g. solution principles descriptions, not specific solutions) should be documented in this clause as soon as possible during the study.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.2 Security area #2: RAN security
| |
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.2.1 Introduction
| Purpose is to study potential attack vectors, vulnerabilities, security and privacy risks, impact and mitigations. This includes the following aspects:
Editor’s Note: To be aligned with TR 38.914 [1] and TR 38.760-2 [2] as 6G RAN study progresses in RAN WGs.
- Radio protocol stack, architecture and procedures
Editor’s Note: Lower layer security is FFS.
Editor’s Note: Examples are FFS.
- Mobility and state transitions within 6G radio
- Mobility between 5G NR and 6G Radio
NOTE: Mobility aspects that affect the core network security context are included in other security areas.
- Interfaces within RAN and between RAN and core network
Editor’s Note: Other aspects are FFS.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.2.2 Security assumptions
| Editor's Note: This clause will document security assumptions related to each security area.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.2.3 Key issues
| Editor’s note: This clause will contain the key issues that need to be addressed by SA3 on each security area. The exact contents are FFS.
5.2.3.y Key issue #2.y: <key issue name>
Editor's Note: Key issues within the security area are not in any particular order but they are added incrementally (y = 1, 2, 3…) when new key issue is identified. 'x' refers to the security area.
5.2.3.y.1 Key issue details
5.2.3.y.2 Security threats
5.2.3.y.3 Potential security requirements
5.2.3.y.4 Interim agreements
Editor's note: This clause will include the principles that are agreed as work progresses for the specific KI#x.y. This may be populated directly or e.g. also when a topic in Area #x gets resolved and a principle is agreed. Where there is consensus, interim agreements pertaining to this key issue (e.g. solution principles descriptions, not specific solutions) should be documented in this clause as soon as possible during the study.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.3 Security area #3: UE to Core Network Security
| |
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.3.1 Introduction
| This security area studies how to establish and manage secure communication(s) between the UE and the Core Network. This includes the following aspects:
Editor's Note: work is to be aligned and in coordination with TR 23.801-01 [4] based on SA2 progress.
Editor's Note: Any potential NAS impact due to the use of AEAD will be based on the conclusions in TR 33.771 [5].
- Security of NAS protocol, architecture and procedures
- NAS Security context management, including mobility
- Interworking between 6GS and 5GS
NOTE: Mobility aspects that are excluded in the RAN security area (i.e., mobility aspects that affect the core network security context) are included here.
Editor’s Note: Other aspects are FFS.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.3.2 Security assumptions
| Editor's Note: This clause will document security assumptions related to each security area.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.3.3 Key issues
| Editor’s note: This clause will contain the key issues that need to be addressed by SA3 on each security area. The exact contents are FFS.
5.3.3.y Key issue #1.y: <key issue name>
Editor's Note: Key issues within the security area are not in any particular order but they are added incrementally (y = 1, 2, 3…) when new key issue is identified. 'x' refers to the security area.
5.3.3.y.1 Key issue details
5.3.3.y.2 Security threats
5.3.3.y.3 Potential security requirements
5.3.3.y.4 Interim agreements
Editor's note: This clause will include the principles that are agreed as work progresses for the specific KI#x.y. This may be populated directly or e.g. also when a topic in Area #x gets resolved and a principle is agreed. Where there is consensus, interim agreements pertaining to this key issue (e.g. solution principles descriptions, not specific solutions) should be documented in this clause as soon as possible during the study.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.4 Security area #4: Security for Core Network, Interconnect and Roaming
| |
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.4.1 Introduction
| Study potential aspects to secure core network communication for the different communication modes within a network and between networks to mitigate threats identified. This is based on the system architecture requirements developed in TR 23.801-1 [6].
Editor's note: This study will be based on agreements documented in TR 23.801-1. In its early phase, prior to interim agreements and conclusions being available, the work in this study shall rely on the Architectural Assumptions and Requirements specified in clause 4 of TR 23.801-1.
This includes the following aspects
• Authentication and authorization between network entities
• Securing the communication
◦ within core network
◦ for interconnect and roaming between networks
• for connection with roaming intermediaries
• Network security credential distribution to enable secure communication in a scalable, efficient, interoperable and resilient manner.
NOTE: This does not include subscription credentials.
Editor’s Note: For roaming, coordination with GSMA is required.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.4.2 Security assumptions
| Editor's Note: This clause will document security assumptions related to each security area.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.4.3 Key issues
| Editor’s note: This clause will contain the key issues that need to be addressed by SA3 on each security area. The exact contents are FFS.
5.4.3.y Key issue #1.y: <key issue name>
Editor's Note: Key issues within the security area are not in any particular order but they are added incrementally (y = 1, 2, 3…) when new key issue is identified. 'x' refers to the security area.
5.4.3.y.1 Key issue details
5.4.3.y.2 Security threats
5.4.3.y.3 Potential security requirements
5.4.3.y.4 Interim agreements
Editor's note: This clause will include the principles that are agreed as work progresses for the specific KI#x.y. This may be populated directly or e.g. also when a topic in Area #x gets resolved and a principle is agreed. Where there is consensus, interim agreements pertaining to this key issue (e.g. solution principles descriptions, not specific solutions) should be documented in this clause as soon as possible during the study.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.5 Security area #5: Subscription Authentication and Authorization
| |
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.5.1 Introduction
| This security area includes the following security aspects related to authentication and authorization between the UE and the 6GS regardless of access type (i.e., 3GPP access and/or non-3GPP access):
Editor’s Note: Whether trusted or untrusted non-3GPP access, or both are in scope is FFS.
-Authentication, key agreement and authorization between the UE and the 6GS.
Editor’s Note: Examples are FFS
Editor’s Note: Other types of authentication is FFS
-Re-authentication between the UE and the 6GS in different conditions of mobility.
-Subscriber identifier privacy.
-Long term credentials storage and processing
Editor’s Note: Authentication between the UE and an external data network (DN) is FFS.
Editor’s Note: Other aspects are FFS
Editor’s Note: clarification of authorization aspects are FFS
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.5.2 Security assumptions
| Editor's Note: This clause will document security assumptions related to each security area.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.5.3 Key issues
| Editor’s note: This clause will contain the key issues that need to be addressed by SA3 on each security area. The exact contents are FFS.
5.5.3.y Key issue #3.y: <key issue name>
Editor's Note: Key issues within the security area are not in any particular order but they are added incrementally (y = 1, 2, 3…) when new key issue is identified. 'x' refers to the security area.
5.5.3.y.1 Key issue details
5.5.3.y.2 Security threats
5.5.3.y.3 Potential security requirements
5.5.3.y.4 Interim agreements
Editor's note: This clause will include the principles that are agreed as work progresses for the specific KI#x.y. This may be populated directly or e.g. also when a topic in Area #x gets resolved and a principle is agreed. Where there is consensus, interim agreements pertaining to this key issue (e.g. solution principles descriptions, not specific solutions) should be documented in this clause as soon as possible during the study.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.6 Security area #6: security and privacy aspects of network exposure
| |
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.6.1 Introduction
| This security area covers the following aspects.
- The security and privacy aspects of the exposure mechanism(s) defined in TR 23.801-01 [4].
Editor’s Note: Other aspects are FFS.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.6.2 Security assumptions
| Editor's Note: This clause will document security assumptions related to each security area.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 5.6.3 Key issues
| Editor’s note: This clause will contain the key issues that need to be addressed by SA3 on each security area. The exact contents are FFS.
5.6.3.y Key issue #1.y: <key issue name>
Editor's Note: Key issues within the security area are not in any particular order but they are added incrementally (y = 1, 2, 3…) when new key issue is identified. 'x' refers to the security area.
5.6.3.y.1 Key issue details
5.6.3.y.2 Security threats
5.6.3.y.3 Potential security requirements
5.6.3.y.4 Interim agreements
Editor's note: This clause will include the principles that are agreed as work progresses for the specific KI#x.y. This may be populated directly or e.g. also when a topic in Area #x gets resolved and a principle is agreed. Where there is consensus, interim agreements pertaining to this key issue (e.g. solution principles descriptions, not specific solutions) should be documented in this clause as soon as possible during the study.
5.x Security area #x: <security area name>
Editor's Note: The study is expected to be divided into several security areas which all have their own key issues and solutions. Security areas are not in any particular order but they are added incrementally (x = 1, 2, 3…) when new area is identified.
5.x.1 Introduction
Editor's Note: Detailed description of the security area
5.x.2 Security assumptions
Editor's Note: This clause will document security assumptions related to each security area.
5.x.3 Key issues
Editor’s note: This clause will contain the key issues that need to be addressed by SA3 on each security area. The exact contents are FFS.
5.x.3.y Key issue #x.y: <key issue name>
Editor's Note: Key issues within the security area are not in any particular order but they are added incrementally (y = 1, 2, 3…) when new key issue is identified. 'x' refers to the security area.
5.x.3.y.1 Key issue details
5.x.3.y.2 Security threats
5.x.3.y.3 Potential security requirements
5.x.3.y.4 Interim agreements
Editor's note: This clause will include the principles that are agreed as work progresses for the specific KI#x.y. This may be populated directly or e.g. also when a topic in Area #x gets resolved and a principle is agreed. Where there is consensus, interim agreements pertaining to this key issue (e.g. solution principles descriptions, not specific solutions) should be documented in this clause as soon as possible during the study.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 6 Solutions
| 6.x Solutions to Security Area #x <security area name>
6.x.y Solutions to Key Issue #x.y <key issue name>
6.x.y.z Solution #x.y.z: <solution name>
Editor's Note: Solutions are added incrementally (z = 1, 2, 3…) when new solution is identified. 'x' refers to the security area, y to the key issue. If a solution addresses multiple key issues, a cross-reference of the solution needs to be provided.
6.x.y.z.1 Introduction
6.x.y.z.2 Solution details
6.x.y.z.3 Evaluation
Editor’s note: Evaluation needs to explain how the solution fulfils each requirement of the key issue.
|
f0d8ac8ba87e5fa314bc150073c98efb | 33.801-01 | 7 Conclusions
| Editor’s note: This clause will contain the overall conclusions made by SA3. The structure of this clause is FFS
Annex A
Attacker Model
A.1 General
Editor's Note: This clause includes an introduction to the attacker model.
A.2 Architecture overview
Editor's Note: This will need to be updated as work progresses in other work groups.
The high-level architecture model of the 6G System is shown in Figure A.2-1 is expected to be based on the high-level architecture of 5G. This architecture model can be used as a basis to introduce some initial attacker model information before the 6G architecture is developed by other 3GPP WGs.
The architecture model includes a set of UEs, a set of Access Networks comprising multiple Access Nodes, a set of Core Networks, several external Data Networks and several Applications.
Editor's Note: Further details are FFS.
Figure A.2-1 High level architecture model
A.3 Attacker Description
Editor's Note: This clause includes an attacker model description.
Annex B
Risk analysis of MAC-CE
Editor’s Note: Structure of annex is FFS.
Editor’s Note: Format of the framework capturing risk analysis is FFS.
Editor’s Note: Methodology for the risk analysis is FFS.
B.1 General
In LTE and 5GNR, security for Control Plane (CP) and User Plane (UP) traffic between the User Equipment (UE) and the base station is fundamentally anchored at the Packet Data Convergence Protocol (PDCP) layer. The risk of Medium Access Control (MAC) layer needs to be analysed. This Annex captures the security and privacy risk analysis of the MAC-CEs from clause 6.1.3 of TS 38.321.
The MAC Control Element (MAC-CE) is a signaling message used at the MAC layer to manage time-critical control functions. For example, MAC-CEs are used for Layer 2 operations, conveying control information for resource management, scheduling, power control, and link maintenance. MAC-CEs were introduced in Release 8 (LTE) and has been expanded in every subsequent release.
Editor’s Note: The alignment of above paragraph with RAN2 is FFS.
Annex <F>:
Change history
Change history
Date
Meeting
TDoc
CR
Rev
Cat
Subject/Comment
New version
2025-10
SA3#124
Initial version
0.0.0
2025-10
SA3#124
S3-253773
Adding Scope to the draft TR
0.1.0
2025-10
SA3#124
S3-253772
Proposal for an Attacker model Annex in the 6G TR 33.801-01
0.1.0
2025-10
SA3#124
S3-253811
Annex mapping of solutions to key issues
0.1.0
2025-10
SA3#124
S3-253812
Adding EN to interim agreements
0.1.0
2025-10
SA3#124
S3-253664
New Security Area on UE to Core Network Security
0.1.0
2025-10
SA3#124
S3-253774
New Security Area on 6G RAN Security
0.1.0
2025-10
SA3#124
S3-253776
Pseudo-CR on Security area Authentication and Authorization
0.1.0
2025-11
SA3#125
S3-254646
Update on Annex B: Mapping of Solutions to Key Issues
0.2.0
2025-11
SA3#125
S3-254650
Updating access-agnostic authentication in Security area #3
0.2.0
2025-11
SA3#125
S3-254654
6G new security area: exposure security
0.2.0
2025-11
SA3#125
S3-254652
New Security Area - Security Architecture
0.2.0
2025-11
SA3#125
S3-254651
Update Security area #3 for adding secondary authentication
0.2.0
2025-11
SA3#125
S3-254647
Pseudo-CR on System overview for the Attacker model
0.2.0
2025-11
SA3#125
S3-254653
New Security Area on Security for Core Network, Interconnect and Roaming
0.2.0
2025-11
SA3#125
S3-254655
Pseudo-CR on key issue related to MAC layer risk mitigation
0.2.0
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 1 Scope
| The present document provides the description and investigation of new AI/ML based use cases, i.e., multi-hop UE trajectory, AI/ML based intra-CU LTM, and other handover enhancements.
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 2 References
| The following documents contain provisions which, through reference in this text, constitute provisions of the present document.
- References are either specific (identified by date of publication, edition number, version number, etc.) or non‑specific.
- For a specific reference, subsequent revisions do not apply.
- For a non-specific reference, the latest version applies. In the case of a reference to a 3GPP document (including a GSM document), a non-specific reference implicitly refers to the latest version of that document in the same Release as the present document.
[1] 3GPP TR 21.905: "Vocabulary for 3GPP Specifications".
[2] 3GPP TS 38.300: "NR; NR and NG-RAN Overall Description".
[3] 3GPP TS 38.401: "NG-RAN; Architecture description".
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 3 Definitions of terms, symbols and abbreviations
| |
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 3.1 Terms
| For the purposes of the present document, the terms given in TR 21.905 [1] and the following apply. A term defined in the present document takes precedence over the definition of the same term, if any, in TR 21.905 [1].
<Void>
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 3.2 Symbols
| For the purposes of the present document, the following symbols apply:
<Void>
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 3.3 Abbreviations
| For the purposes of the present document, the abbreviations given in TR 21.905 [1] and the following apply. An abbreviation defined in the present document takes precedence over the definition of the same abbreviation, if any, in TR 21.905 [1].
<Void>
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4 Use cases and Solutions
| |
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4.1 Multi-hop UE trajectory
| |
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4.1.1 Use case description
| Editor’s Note: Capture the description of use case
In Rel-18, the cell-based UE trajectory prediction is limited to the first-hop target NG-RAN node.
Multi-hop predicted UE trajectory across gNBs consists of a list of cells belonging to gNBs where the UE is expected to connect and these cells are listed in chronological order.
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4.1.2 Solutions and standard impacts
| Editor’s Note: Capture the solutions for the use case, including potential standard impacts on existing Nodes, functions, and interfaces
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4.1.2.1 Locations for AI/ML Model Training and AI/ML Model Inference
| The following solutions are considered for supporting multi-hop UE trajectory:
- AI/ML Model Training is located in the OAM and AI/ML Model Inference is located in the gNB.
- AI/ML Model Training and AI/ML Model Inference are both located in the gNB.
In case of CU-DU split architecture, the following solutions are possible:
- AI/ML Model Training is located in the OAM and AI/ML Model Inference is located in the gNB-CU.
- AI/ML Model Training and Model Inference are both located in the gNB-CU.
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4.1.2.2 Input data of multi-hop UE trajectory
| To predict the multi-hop UE trajectory, a gNB may need the following information as input data:
From the UE:
- UE measurement report related to serving cell and neighbouring cells, e.g., RSRP, RSRQ, SINR
- UE Mobility History Information
From the neighbouring RAN nodes:
- UE Mobility History Information
From the local node:
- Multi-hop predicted UE trajectory
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4.1.2.3 Output data of multi-hop UE trajectory
| The following information can be generated as output:
- Multi-hop predicted UE trajectory, including a list of cells that UE is expected to connect to in chronological order and the associated expected time UE stays in the cell
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4.1.2.4 Feedback of multi-hop UE trajectory
| To optimize the performance of multi-hop UE trajectory prediction, the following feedback can be considered to be collected from gNBs:
- Measured UE Trajectory collected at each individual gNB
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4.1.2.5 Standard impact
| Multi-hop UE trajectory prediction is transferred to the target NG-RAN nodes over the Xn interface via the Handover Preparation procedure. For the subsequent handovers, multi-hop predicted UE trajectory includes a list of cells where the UE is expected to connect to in chronological order.
For the subsequent handovers, Data Collection Reporting Initiation procedure needs to be in place between the source gNB and a subsequent gNB to request the reporting of the measured UE trajectory. Each target gNB directly transmits the collected measured UE trajectory back to the source gNB.
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4.2 AI/ML assisted Intra-CU LTM
| |
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4.2.1 Use case description
| Editor’s Note: Capture the description of use case
L1/L2 Triggered Mobility (LTM) is specified in TS 38.300 [2].
Intra-CU LTM is specified in TS38.401[3].
AI/ML can be used to optimise Intra-CU LTM procedures, e.g., to enhance Network and UE performance, optimize resource allocation and reduce mobility failures.
AI/ML assisted L3 and L1 measurements based intra-CU LTM with inference in gNB-CU are studied.
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4.2.2 Solutions and standard impacts
| Editor’s Note: Capture the solutions for the use case, including potential standard impacts on existing Nodes, functions, and interfaces
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4.2.2.1 Locations for AI/ML Model Training and AI/ML Model Inference
| For CU-DU split architecture, the following solutions are possible:
- AI/ML Model Training is located in the OAM and AI/ML Model Inference is located in the gNB-CU.
- AI/ML Model Training and Model Inference are both located in the gNB-CU.
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4.2.2.2 Input data of AI/ML based Intra-CU LTM
| Editor’s Note: To be updated
For AI/ML optimization of intra-CU LTM the following information can be considered as input data:
- L3 measurement results
- UE mobility history
- Measured/Predicted radio resource status per cell/SSB area
- Measured/Predicted cell-based UE trajectory
- Historical UE’s candidate LTM cell and beam list(s)
- Measured TA values
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4.2.2.3 Output data of AI/ML based Intra-CU LTM
| For AI/ML optimization of Intra-CU LTM the following information can be considered as output data:
- Candidate cell and beam for LTM HO Preparation- Target cell and beam selection for cell switch command
- Cell(s) and beam(s) for early synchronization
- Predicted TA value(s) for early UL synchronization
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4.2.2.4 Feedback of AI/ML based Intra-CU LTM
| The following data is required as feedback data for intra-CU LTM:
- LTM target cell
- Measured TA value
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4.3 Other handover enhancements
| Editor’s Note: Identify other handover enhancements via AI/ML, e.g., inter-CU LTM
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 4.3.1 AI/ML assisted inter-CU LTM
| Inter-CU LTM is specified in TS 38.300 [2].
AI/ML may be used to optimise inter-CU LTM procedures, e.g., for candidate cell selection.
Editor’s Note: If applicable, use agreements for Intra-CU LTM as baseline for inter-CU LTM．
|
6dfc46bcb9bf4ef0655f25761d065721 | 38.745 | 5 Conclusion
|
Annex A:
Change history
Change history
Date
Meeting
TDoc
CR
Rev
Cat
Subject/Comment
New version
2025-10
RAN3#129-bis
R3-256546
Skeleton for TR38.745 v0.0.0
0.0.0
2025-10
RAN3#129-bis
R3-257331
TR 38.745 for Study on AI/ML for NG-RAN Phase 3
0.0.1
2025-10
RAN3#129-bis
R3-257331
TR38.745 v0.1.0 for Study on Artificial Intelligence (AI)/Machine Learning (ML) for NG-RAN Phase 3
0.1.0
2025-11
RAN3#130
R3-258086
TR38.745 v0.1.1 for Study on Artificial Intelligence (AI)/Machine Learning (ML) for NG-RAN Phase 3
0.1.1
2025-11
RAN3#130
R3-258878
TR38.745 v0.2.0 for Study on Artificial Intelligence (AI)/Machine Learning (ML) for NG-RAN Phase 3
0.2.0
2025-12
RAN#110
RP-253139
TR 38.745: Study on Artificial Intelligence (AI)/Machine Learning (ML) for NG-RAN Phase 3
1.0.0
|
a42d942a68c46b17b4d16e1f71329da3 | 38.765 | 1 Scope
| The present document is intended to capture the output of study item for "Study on Integrated Sensing And Communication (ISAC) for NR" [2]. CP-OFDM is considered as baseline waveform. The purpose of this TR is to document the following investigations for NR ISAC.
- Performance evaluation of gNB-based mono-static sensing for UAV use case
- Performance metrics and related performance objectives
- Evaluation assumptions
- Evaluation results
- Measurements and quantization
- Procedures and signalling between RAN and CN to support ISAC
- Network architecture for gNB-based mono-static sensing for UAV use cases
|
a42d942a68c46b17b4d16e1f71329da3 | 38.765 | 2 References
| The following documents contain provisions which, through reference in this text, constitute provisions of the present document.
- References are either specific (identified by date of publication, edition number, version number, etc.) or non‑specific.
- For a specific reference, subsequent revisions do not apply.
- For a non-specific reference, the latest version applies. In the case of a reference to a 3GPP document (including a GSM document), a non-specific reference implicitly refers to the latest version of that document in the same Release as the present document.
[1] 3GPP TR 21.905: "Vocabulary for 3GPP Specifications".
[2] 3GPP RP-252819: "Revised SID: Study on Integrated Sensing And Communication (ISAC) for NR".
|
a42d942a68c46b17b4d16e1f71329da3 | 38.765 | 3 Definitions of terms, symbols and abbreviations
| |
a42d942a68c46b17b4d16e1f71329da3 | 38.765 | 3.1 Terms
| For the purposes of the present document, the terms given in TR 21.905 [1] and the following apply. A term defined in the present document takes precedence over the definition of the same term, if any, in TR 21.905 [1].
example: text used to clarify abstract rules by applying them literally.
|
a42d942a68c46b17b4d16e1f71329da3 | 38.765 | 3.2 Symbols
| For the purposes of the present document, the following symbols apply:
<symbol> <Explanation>
|
a42d942a68c46b17b4d16e1f71329da3 | 38.765 | 3.3 Abbreviations
| For the purposes of the present document, the abbreviations given in TR 21.905 [1] and the following apply. An abbreviation defined in the present document takes precedence over the definition of the same abbreviation, if any, in TR 21.905 [1].
CPI Coherent Processing Interval
STX Sensing Transmitter
SRX Sensing Receiver
UAV Uncrewed Aerial Vehicles
TRP Transmission-Reception Point
LCS Local coordinate system
GCS Global coordinate system
|
a42d942a68c46b17b4d16e1f71329da3 | 38.765 | 4 Performance metrics
| Editor’s note: This section is to capture the definition of performance metrics for the evaluation, and if agreed, the targeted KPI values
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.