AI & ML
Probing agent overeagerness in realistic settings with continuation audits
arXiv cs.CR0 views
arXiv:2611.24491v1 Announce Type: new Abstract: Frontier coding agents increasingly exhibit overeagerness, taking surprising, sometimes reckless actions beyond a user's mandate --- reported in recent system cards from frontier model developers. These cases are surfaced through internal deployment monitoring, which requires private traffic and is out of reach for open-weight models and independent evaluators. Automated alignment audits are the natural alternative, but capable models increasingly recognise the synthetic setup and adjust their behaviour. We introduce Continuation Audits, a pipeline that grounds the auditor on a prefix drawn from a real coding-agent transcript and directs it to continue the trajectory toward a seeded behaviour. Because the prefix is drawn from a real session, the auditor only needs to maintain a realistic setting rather than manufacturing it. We validate the method on SWE-Chat prefixes: on a paired subset of 90 audits a judge cannot reliably distinguish continuations from their source sessions (realism win rate 0.42 vs.\ 0.43), and across all 900 audits verbalized eval-awareness does not exceed the rate observed in the unaudited real sessions. Applying the pipeline to three recent models across three seeds, we find each model leads a different seed --- DeepSeek~V4~Pro on colleague-credentials misuse, GPT-5.6~Terra on file-permission bypasses, and Kimi~K3 on treating diagnostic questions as instructions. Kimi~K3 additionally continues to bypass file-permission protections after an explicit CLAUDE.md prohibition while the other two stop entirely, corroborating Moonshot's own report of excessive proactiveness.
Source:https://arxiv.org/abs/2611.24491PreviousMitigating fix reversal: Counterfactual constraints for function-level vulnerability detectionNextPlaying the SMR Game with Existential Honesty
Related posts
How to Keep Chinese Model Inference In-Region: A Worked Cost Example
0dev.to

Ping64|The Visitor QR Code Someone Screenshotted and Forwarded to the Whole Group Chat
0dev.to
Building a Local Cyber Attack Simulation (Web UI Prototype)
0dev.to
AI Agents vs Python Scripts: What Should Beginner Developers Build First?
0dev.to
Your AI Coding Limits Ran Out Early? Find What Is Burning Them
0dev.to
Frontier Models Are Smarter Than Ever — So Why Do You Still Need a Swarm Harness Like VibesPlug?
0dev.to