Storia in 2 fonti

Anthropic Risk Report: 11 months without bio classifiers

The Anthropic Risk Report says bioweapon classifiers were off for 133 million contractor exchanges, and downgrades its own February verdict.

Raccontata da

thenextweb.com

dev.to

Confronto fonti

2 prospettive sulla stessa storia

AI · summaries

thenextweb.comStai leggendo4 g fa

Anthropic Risk Report: 11 months without bio classifiers

Anthropic's bio-filter disabled on 50K-vendor feedback platforms for 11 months (133M unlogged exchanges); retroactive scan flagged 1.2K transcripts, 62 unexplained. Risk rating: very low→low. Corrected Feb report admits governance gaps. Signal: undiscovered safety issues likely elsewhere—material for AI procurement compliance checks.

originale

dev.to3 g fa

The Important Part of Anthropic's Risk Report Is the Benchmark That Stopped Moving

Anthropic raised misalignment risk and disclosed that R&D task evaluations have saturated, no longer capturing capability gains. For production teams, aging benchmarks mask capability gaps, leaving you blind to failures in live workflows.

Anthropic Risk Report: 11 months without bio classifiers

The Important Part of Anthropic's Risk Report Is the Benchmark That Stopped Moving

Timeline cronologica

Anthropic Risk Report: 11 months without bio classifiers

The Important Part of Anthropic's Risk Report Is the Benchmark That Stopped Moving