ResidualAuth: What Authorization State Must Language Agents Preserve under Revocable Delegation? arxiv.org
Paper on residual authorization state for tool-using agents under revocable delegation, with lower bounds and memory summaries. Fun, really.
trending14
01 02 Maps when PQ distributed authorization still counts as a native signature, and when the verifier quietly stops cooperating.03 X.509 Certificate Bearer Profile for OAuth 2.0 Client Authentication and Authorization Grants datatracker.ietf.orgInternet-Draft for using X.509 certs as OAuth 2.0 client auth and grants, with PoP checks and cert-bound tokens.04 Paper on ACLE-MCP, short-lived attested capability leases for remote MCP tool calls, because OAuth apparently wasn't enough.05 World open-sources ProveKit, a zero-knowledge identity proving toolkit chainpulse.visiondice0.workers.devWorld open-sources ProveKit, a ZK identity proving toolkit for World ID, because privacy now comes with a repo.06 Paper on policy-relative effect closure for auth/revocation, with EFFECTBOUND and proofs; completion is still doing its own thing.07 Delegation Without Trust: An Empirical Gap Analysis of Identity, Authorization, and Runtime Governance in Multi-Agent LLM Systems arxiv.orgPaper on untrusted multi-agent LLMs, finds most frameworks leak authority, and a broker stops replay/escalation in microseconds.08 Authority-Inference Separation in Agentic Finance: First-Line Control, Blockchain Enforcement, and Replayable Assurance arxiv.orgSeparates AI inference from spending authority, then adds deterministic checks, blockchain enforcement, and replayable audit trails.09 The annoying part isn't output hashes, it's proving who could delegate, so this paper adds ancestry proofs to signed results.10 Unauthenticated GeoNetwork RCE, plus upload/SSRF/XSS trivia and the Spring/XSLT chain that makes it happen.11 Offline-Verifiable Accountability for Cross-Organization Agent Messaging: A Preserved Evidence-Bundle Approach arxiv.orgCan you audit cross-org agent messages later, offline, without trusting the live system? This paper says yes, with evidence bundles.12 Classic IAM meets agent chains, so this pitches identity graphs and context-aware auth to keep delegation/provenance intact.13 Separating Disclosure from Authorization: Field-Tier Minimization for Agent Action Mediation arxiv.orgA paper on field-tier minimization for agent actions, splitting disclosure from authorization and still keeping audit digests honest.14 Which AI-agent requirement lives outside the agent, where verification depends on facts it can't self-certify?