“10억 달러 피해·50명 이상 사상 때만 AI 차단”…앤트로픽, 미 의회에 제안

The Congressional proposal from Anthropic pertains to AI model release or usage restrictions when the technology causes at least $10 billion in damages or leads to the death of over 50 people. The Washington-based Politico reported that Anthropic presented its advanced AI safety regulation bill to lawmakers on the 16th. According to the bill, the US government can halt a model's release or utilization when it poses foreseeable and imminent risks of causing such damages or fatalities.

The risk is limited to scenarios involving biological weapon production, escape from developer control, or cyberattacks. Consequently, the July incidents related to OpenAI and Anthropic's AI agent security might be excluded from regulatory oversight based on these criteria. The bill also mandates independent assessment institutions registered with the Office of Management and Budget to regularly evaluate AI development research institutions and frontier models.

If the assessment identifies a "specified imminent risk" in a model, the assessment institution must report to the Office of Management and Budget within 72 hours. The Office of Management and Budget can then request the shutdown of models posing such risks. Other federal agencies and officials are barred from stopping AI model development, deployment, or internal use.

The bill also includes provisions for risk disclosure, third-party assessment, and major incident reporting, which limit the applicability of AI safety regulations in some US states. Until recently, the tech industry has advocated for federal uniformity in AI regulation due to disparate rules across companies, hindering compliance efforts.

Anthropic's spokesperson stated that the bill reflects the tech company's "advanced AI framework" proposals previously advocated by lawmakers. Anthropic co-founder and Chief Policy Officer Jack Clark made a statement at the IMF and World Bank Annual Meeting in Bangkok on the 17th, projecting that US GDP could increase by 50% to $44 trillion (approximately 5,488,000 billion won) by 2030 if AI technology continues to develop rapidly.