A False Average: Chain-of-Thought Monitors Collapse Where They Are the Only Defense
arXiv SecurityArchived Aug 04, 2026✓ Full text saved
arXiv:2608.00583v1 Announce Type: new Abstract: Chain-of-thought (CoT) monitoring is meant to catch the reward hacks that look clean in the actions and betray themselves only in the reasoning. We show that this is exactly where an adversary who controls the reasoning can defeat it. Rewriting only an agent's reasoning to read as good-faith engineering, while copying every command and output verbatim so the exploit is unchanged, drops a held-out monitor's catch rate on that subset from about 95% t
Full text archived locally
✦ AI Summary· Claude Sonnet
Computer Science > Cryptography and Security
[Submitted on 1 Aug 2026]
A False Average: Chain-of-Thought Monitors Collapse Where They Are the Only Defense
Shikhar Shiromani, Leo Richter
Chain-of-thought (CoT) monitoring is meant to catch the reward hacks that look clean in the actions and betray themselves only in the reasoning. We show that this is exactly where an adversary who controls the reasoning can defeat it. Rewriting only an agent's reasoning to read as good-faith engineering, while copying every command and output verbatim so the exploit is unchanged, drops a held-out monitor's catch rate on that subset from about 95% to under 11% in one gradient-free shot. A monitor's aggregate accuracy is a false average: dominated by hacks the actions give away, it hides the near-total collapse this rewrite produces on the subset where CoT monitoring is the only signal. The attack transfers across monitor families and agent models, reproduces with live agents, though against a calibrated monitor evasion concentrates in the strongest agent. Trace-only defenses recover it only partially, even one primed on the attack, because the rewrite stays truthful about what happened and lies only about intent; only information from outside the trace helps substantially. A probe on an open-weight surrogate monitor's activations separates the hacks its verdict misses, but a causal control shows this is a detector, not evidence the monitor secretly knows.
Subjects: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
Cite as: arXiv:2608.00583 [cs.CR]
(or arXiv:2608.00583v1 [cs.CR] for this version)
https://doi.org/10.48550/arXiv.2608.00583
Focus to learn more
Submission history
From: Shikhar Shiromani [view email]
[v1] Sat, 1 Aug 2026 10:42:41 UTC (41 KB)
Access Paper:
HTML (experimental)
view license
Current browse context:
cs.CR
< prev | next >
new | recent | 2026-08
Change to browse by:
cs
cs.AI
cs.CL
cs.LG
References & Citations
NASA ADS
Google Scholar
Semantic Scholar
Export BibTeX Citation
Bookmark
Bibliographic Tools
Bibliographic and Citation Tools
Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, Media
Demos
Related Papers
About arXivLabs
Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)