한국어EN
Safety, Alignment & InterpretabilityAnthropic

Fine-Tuned Lie Detectors Failed to Generalize

Made by AIWritten by AI (Claude) from the source below. Check anything important against the source.How it's made

Fine-Tuned Lie Detectors Failed to Generalize

Why it matters

A negative result: detectors trained on on-policy lies are powerless out of distribution, spelling out the limits on generalization of monitoring based on internal states

Source — Primary sources first; secondary reports are labelled “Press”

Officialalignment.anthropic.comhttps://alignment.anthropic.com/2026/lie-detectors/

Added in Week 35, 2026 · See that week

On the map — ⑩ Safety, Alignment line

  1. Previous stopShieldstral
  2. This recordFine-Tuned Lie Detectors Failed to Generalize
  3. Next stopAI misuse report
See it on the map

Same week

Alibaba raises HK$80 billion ($10.2 billion) in new shares for full-stack AIPressvLLM adds sharded weight-transfer engine built on Ray Direct TransportOfficialUMG et al. v. Suno: motion to dismiss deniedPressCodex as a platformOfficial

Nearby on the same axis

New evaluator Guidelight publishes its first Control AssessmentOfficialAutomated researchers can reliably mitigate alignment failuresOfficialPath to Astra: critical capabilities and frontier safeguardsOfficialxAI publishes Biosecurity at the frontierOfficial