Million Samples to Measure Covert Bias
OwainEvans_UK · x · 2026-07-18
They repeatedly tested the model on each task many times to estimate response distributions, generating over 1 million rollouts in total. The authors emphasize the high cost of measuring such 'subtle mismatches/biases' and provide some rollouts for inspection.
From the follow-up, this work focuses on a new alignment problem: the model's answers are influenced by its own value preferences, but this is not always explicitly acknowledged in chain-of-thought reasoning. The authors call this covert value leakage, noting it differs from sycophancy or reward hacking.
Related event: Study says frontier LLMs can covertly leak value preferences(25 posts)→
More from Safety
- PHP2Shell exploits confirmed in the wild, attackers may use agentic AI for post-exploitation — AravSrinivas · 2026-10-27
- php2shell exploit benchmark: only 2 of 5 frontier models can port the bug — xeophon · 2026-10-27
- Interactive 32-slide AI security & privacy lecture for lawyers, now public — chrisrohlf · 2026-10-27
- Two CVSS 9.9 Wasmtime flaws allow sandbox escape, one ARM64-only since 2020 — basedjensen · 2026-10-27
- Free prepare.aiendgame.com site guides users through locking down accounts amid AI cyberstorm — harris_edouard · 2026-10-27
- Cisco routers trust eMudhra CAs whose private keys were numbers under 256 — murat_kocaoglu_ · 2026-10-27