Echoes within the Reasoning: Stealthy and Effective Watermarking via Chain of Thought
2026-05-29T07:23:32Z•9c64042e365bb4fd97b5d72bc90744acde47d47d4d0e8a45bfeca0cc0ad4c1ce
Ethereum validator privacyGEO-BenchLLM watermarkingRobust Subspace RegistrationTOCTOU mitigationTor pushadversarial MLagent securityauthority confusionautonomous driving vulnerabilitiesbenchmarkingchain-of-thoughtdelegated privilege promotiondifferential privacye-valuesprompt injectionranking manipulationruntime authorizationunix-ctfvision-language-action modelswatermark verification
What happened
Collection of recent security- and privacy-focused CS papers (arXiv 2026-05-29) covering vulnerabilities, defenses, and benchmarks for modern ML/agent systems and infrastructure. Key contributions include BiCoT — a stealthy Chain-of-Thought watermarking scheme with a Robust Subspace Registration black-box verifier; AIRGuard — a runtime authority-control layer that dramatically reduces agent-side attack success; a large-scale measurement showing ~1% of real-world resumes contain prompt injections; ReasonBreak — demonstrations of high success rates (up to 89% reasoning, 72% trajectory) for input
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- arxiv_cs_cr
- Record identifier
- 9c64042e365bb4fd97b5d72bc90744acde47d47d4d0e8a45bfeca0cc0ad4c1ce
- Enrichment time
- 2026-05-29T07:23:32Z
- AI-assisted enrichment
- Yes
This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.