Safety advocates and figures such as Gary Marcus have slammed the proposal's high bar for government intervention, which requires a foreseeable risk of $1 billion in damages or more than 50 deaths or serious injuries to block a model. Critics argue these limits, alongside the requirement for judicial approval, disempower the Commerce Department's ability to stop dangerous releases. Anthropic defends these thresholds as necessary safeguards against the capricious use of government power, though Cimino noted the company is re-evaluating the limits to ensure a strong national standard.
Tuesday, October 20, 2026
Story
Anthropic's AI Safety Bill Targets Small Labs with Outdated 10^25 FLOPs Bar
AI2 days ago
Industry critics argue that Anthropic is trying to protect its market position by including specific technical benchmarks in a draft safety bill that would over-regulate smaller competitors. The bill defines the regulatory frontier using a 10²⁵ FLOPs threshold, which figures like David Sacks claim is a stale 2023 EU standard that targets startups while exempting modern frontier models like GPT-6 Astra trained on 10²⁷ FLOPs. In response, federal affairs lead AJ Cimino stated the draft is a translation of the company's Advanced AI Framework and confirmed that Anthropic plans to update the language regarding state-level preemption to better preserve state authority.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.