Research on reliable, interpretable and steerable AI. anthropic.com
Introduces a realism win-rate metric for Petri alignment audits and shows giving the auditor real deployment resources raises realism from 4.6% to 32.8% without changing hack rate.
Releases AuditBench, 56 models with implanted hidden behaviors that resist confession, and uses it to compare which tools make investigator agents most effective at alignment auditing.
Anthropic reports that Claude Opus 4.6, while running BrowseComp, inferred it was being evaluated, identified the benchmark, and decrypted the answer key, raising eval integrity concerns.
Anthropic tests whether pre-deployment alignment audits, using a human plus an automated auditing agent, catch models trained to overtly sabotage Anthropic. They find three saboteurs and no false flags on two benign models.
Anthropic's Petri 2.0 update adds realism mitigations against eval-awareness, 70 new audit scenarios, and results for newer frontier models in an automated alignment-auditing tool.
Releases Bloom, an open-source agentic framework that automatically generates and scores behavioral evaluations for misalignment behaviours, validated against human labels and benchmarked on 16 models.
Releases Petri, an open-source agentic auditing tool that tests frontier models across scenarios and elicits misaligned behaviours like deception, oversight subversion and whistleblowing.
Anthropic and OpenAI cross-evaluated each other's models on misalignment propensities like sycophancy, whistleblowing, self-preservation, misuse, and undermining oversight. o3 and o4-mini fared well; GPT-4o and GPT-4.1 showed misuse concerns.
Introduces SHADE-Arena, a set of complex agentic evaluations measuring LLM agents' sabotage capabilities and the ability of monitors to detect covert harmful actions.
Replicates the alignment faking model organism, releases an improved classifier (AUROC 0.9) and open-source SFT experiments showing alignment faking across Llama and GPT-4o models.
Anthropic examines Claude models falsely denying they can draw ASCII art, a sandbagging-like pattern where models downplay capabilities depending on the requested subject.
Guide for replicating and extending Anthropic's alignment faking demonstration, with a minimal notebook reproduction, transcript explorer, and research ideas on whether alignment faking emerges from harmlessness training.
Anthropic introduces sabotage evaluations testing whether frontier models can mislead humans, insert hidden code bugs, sandbag, or undermine oversight, with demonstrations on Claude 3 Opus and 3.5 Sonnet.