Ericsson and Fraunhofer propose IETF AI agent auditing architecture

Standardizing how autonomous agents report their actions is critical as streaming platforms increasingly deploy AI for dynamic ad insertion, content moderation, and personalized encoding. By linking user intent to final API execution, this architecture provides the accountability necessary for financial transactions and data sharing that traditional logs fail to capture. For the broader ecosystem, this move toward interoperable auditing reduces vendor lock-in by allowing third-party auditors to verify compliance across different cloud and tool providers. Watch for the development of the WI-1-1 delegation-chain record format to see how identity is preserved across complex, multi-agent service calls.

Ericsson has been building a broader portfolio of AI trust and automation standards work that contextualizes this IETF draft. In early 2025, Ericsson joined the AI-RAN Alliance alongside NVIDIA and SoftBank to develop AI-native radio access network architectures, signaling the company's intent to embed autonomous decision-making into critical network infrastructure where auditability becomes a compliance requirement. Mirja Kuehlewind, one of the draft's authors, has been active in IETF transport and security working groups, and her involvement in this proposal reflects Ericsson's strategic interest in establishing governance frameworks before autonomous agents become deeply embedded in telecom and streaming service delivery pipelines.

The regulatory and standards landscape around AI agent accountability is tightening across multiple bodies. The EU AI Act, which entered into force in August 2024 with phased enforcement beginning in 2025, requires providers of high-risk AI systems to maintain detailed logging and traceability records that map closely to the delegation-chain concepts in this draft. Meanwhile, the IETF's RATS (Remote ATtestation procedureS) working group, which this proposal builds upon, published its attestation architecture standard as RFC 9334 in January 2023, establishing the foundational trust model that the new draft extends to multi-agent interaction scenarios. Henk Birkholz, the second author from Fraunhofer SIT, has been a key contributor to RATS and SCITT (Supply Chain Integrity, Transparency, and Trust), positioning this draft as a natural evolution of supply-chain verification techniques applied to AI agent behavior.

On the technical side, the SCITT framework that underpins this proposal has seen concrete implementation progress. The IETF SCITT working group advanced its transparency log architecture through interoperability testing at IETF 122 in March 2025, demonstrating that signed statements can be verified across independent transparency services without requiring a single trusted operator. This matters for streaming and media workflows where multiple vendors handle content processing, ad decisioning, and delivery optimization. The audit architecture proposed by Ericsson and Fraunhofer SIT would allow a platform operator to cryptographically verify that an autonomous agent's ad insertion decision, for example, was authorized by a specific user intent and executed within defined parameters, creating a tamper-evident record suitable for regulatory review or contractual dispute resolution.