Safety, Alignment & InterpretabilityAnthropic (Benton et al.)Anthropic

Sabotage Evaluations for Frontier Models

Sabotage Evaluations for Frontier Models

Why it matters

Formalized a model's capacity to undermine human oversight, evaluation and decision-making as a distinct risk category

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

Paperarxiv.orghttps://arxiv.org/abs/2410.21514

Added in Week 35, 2026 · See that week

Same week

π0 — flow matching, up to 50 Hz, 8 robot typesOfficialOpen Source AI Definition 1.0Official

Nearby on the same axis

Google DeepMind publishes SynthID-Text in NaturePaperRSP v2.0OfficialAlignment Faking in LLMsOfficialDeliberative AlignmentPaper