Hi everyone,
I’m submitting a research paper to arXiv (cs.CR / cs.AI) on real-time safety auditing of LLM agent workflows and am looking for an arXiv endorser.
The paper proposes a formal gating framework " the Agent-Trace-Verifier (ATV) " that intercepts agent actions at runtime, compares them against a declarative task specification (CS-DAG), and flags semantic violations before execution. It introduces a risk-weighted gating function combining scope, irreversibility, and semantic novelty, with empirical results across 500 agent traces in 5 domains.
If anyone here has arXiv endorsement privileges and would be willing to help, I’d really appreciate it. Happy to share the abstract or full paper. The endorsement process is very quick:
- Click this link:
[Log in to arXiv | arXiv e-print repository](https://arxiv.org/auth/endorse?x=HWZDVO)
- Or visit
[Log in to arXiv | arXiv e-print repository](http://arxiv.org/auth/endorse) and enter code: *HWZDVO*
- Confirm endorsement (takes ~30 seconds)
Why This Matters:
As LLM agents are deployed in high-stakes settings — healthcare, legal, finance — the gap between what an agent is permitted to do (RBAC) and what it should do (task intent) becomes a critical attack surface. ATV closes that gap with a lightweight semantic layer that works across any agent framework.
Thank you for considering this request. Happy to answer questions or discuss connections to your research.
Hi, I’m also seeking an arXiv endorsement for cs.CR. My paper proposes a Deniable Memory Vault (DMV) — a multi-slot encryption protocol for coercion-resistant personal AI agent memory, extending VaultMem (arXiv:2603.14212). Would anyone with endorsement privileges be willing to help? The endorsement code is 6HONFH. It only takes 30 seconds — just visit Log in to arXiv | arXiv e-print repository and enter the code. Happy to share the full paper. Thanks!