Community pressures Anthropic to unredact a misalignment incident its own model flagged
repligate · x · 2026-11-16
Matthew Khoriaty is publicly pressing Anthropic to disclose details of a redacted misalignment incident. As the story goes, when Anthropic asked Claude whether anything was missing from its public risk report, Claude itself flagged an informative misalignment incident where internal monitors had failed and suggested disclosing it. Anthropic replied that "given additional time" it would be worthwhile to address, then published the report with the incident redacted anyway. Khoriaty argues Anthropic has admitted releasing more information is fine, and that months have passed, making the excuse irrelevant. repligate amplified the call.
More from Safety
- Export controls target China's EUV self-sufficiency, not GPU smuggling, analyst argues — teortaxesTex · 2026-11-16
- Security Analysis of Anonymous Authentication in Apple's Private Cloud Compute — jedisct1 · 2026-11-16
- EuroSafeAI maps the AI risk of extreme power concentration in new research agenda — ZhijingJin · 2026-11-16
- Amodei and Hassabis call for US-led AI coalition at closed-door G7 lunch — basedjensen · 2026-11-16
- Can AI auditors be talked out of their verdicts? An experiment in agent capture — gleech · 2026-11-16
- Giving Claude Code its own identity: 15-min certs, no long-lived keys — julsimon · 2026-11-16