24 items on this day
An individual draft laying out architectural requirements for hosting AI agents on the Internet, addressing addressing, identity, capability and discovery.
Proposes the Agent Registration and Discovery Protocol (ARDP), specifying how agents are located and how they declare their capabilities.
Defines trust, identity and verifiable provenance for agent-to-agent interaction, giving A2A exchanges an auditable identity layer.
SAIP specifies a Signed Agent Identity Protocol that cryptographically attests agent identity — answering "how do I know which agent I'm talking to".
CAAM proposes a contextual agent authorization mesh that binds authorization decisions to invocation context rather than static tokens.
Combines network digital twins with agentic AI to propose an architecture for AI-driven network operations — the closest to carrier-network practice this week.
Execution Context Tokens for distributed agentic workflows, carrying and validating execution context across cross-domain agent invocations.
Packages agent actions as SCITT-registerable "action capsules", making each agent action auditable and verifiable after the fact.
A SPRING WG draft defining SRv6-based redundancy protection, adding a new protection mechanism for high-availability services over SRv6.
IVY WG advances a YANG data model for inventory topology mapping — foundational for programmable network inventory and agent-gateway resource catalogs.
New RFC specifying distribution of the SRv6 locator via DHCPv6, automating the most tedious part of SRv6 deployment at carrier scale.
OASIS approves Universal Business Language V2.5 as a standard, giving machine-exchangeable business documents a unified schema.
A contract-centered runtime architecture that makes agent capabilities and obligations explicit, targeting the manageability problem of agentic runtimes.
An analytical framework of risks and controls for deploying multi-agent systems — usable as a pre-deployment risk checklist.
FaulT-Bench benchmarks network troubleshooting LLM agents under unreliable user input — a rare agent benchmark for network operations.
Shows how tool outputs become injected commands, arguing to separate action induction from runtime authorization — echoing IETF agent-authorization drafts.
WikiSkill compiles agent experience into persistent knowledge for skill evolution, tackling how agents accumulate and reuse experience.
Anthropic proposes MHS as a unified interface standard bridging AI and the physical world — a vendor-defined interface worth watching.
DeepMind ships Gemini Omni 1.1 Flash with finer developer control over generation behavior.
Argues the real enterprise AI risk is not autonomous agents themselves but the complexity between them, and governance should follow.
Huawei Cloud presents an AI-Ready data infrastructure stack, repositioning the data layer from serving data workloads to serving agents.
Discusses why the agent era is prompting attempts to "rebuild Google" — agents are moving the entry point for information retrieval onto themselves.
Open-source agent skill collection for scientific research, turning research workflows into reusable agent skills.
An open-source multi-agent project from Tsinghua MAIC trending on GitHub — a reference implementation for multi-agent collaboration.
This brief was AI-assisted in collection and summarization, then human-reviewed before publication.