Anthropic plans to revise parts of its AI safety discussion draft. Anthony Cimino, who leads the company’s federal affairs team, announced changes on October 18 after public criticism.
The revised language should more clearly preserve states’ role in independent evaluation and model shutdowns, he said. Cimino also argued that risk thresholds were due for another look.
He outlined possible stronger incident-reporting requirements and further rules for evaluators. His response supplied neither a final revised text nor evidence of implementation.
Palisade Research’s Jeffrey Ladish countered that a proposal must address recursive self-improvement risks. His warning concerns a feared development, rather than an observed takeover of control.