Latest / GNN Daily Breaking News / Anthropic quietly pitches its own AI safety bill to lawmakers

Transcript

Anthropic quietly pitches its own AI safety bill to lawmakers · GNN Daily Breaking News

• 0:00
  Speaker 1: Artificial Intelligence Laboratory Anthropic has privately circulated a confidential legislative
• 0:06
  Speaker 1: proposal to United States lawmakers aimed at establishing a national
• 0:10
  Speaker 1: regulatory framework for frontier artificial intelligence models. According to investigative
• 0:16
  Speaker 1: reporting by Politico, the draft legislation outlines formal safety testing
• 0:22
  Speaker 1: requirements and independent evaluation protocols for advanced developers. while simultaneously
• 0:28
  Speaker 1: establishing exceptionally high legal bars for executive branch intervention. Under
• 0:33
  Speaker 1: the proposed terms, federal regulators would be strictly prohibited from halting, suspending,
• 0:39
  Speaker 1: or restricting the deployment of powerful frontier software systems unless
• 0:44
  Speaker 1: those models pose a clear and imminent threat of causing
• 0:48
  Speaker 1: catastrophic societal harm. The draft bill explicitly restricts emergency federal
• 0:53
  Speaker 1: intervention to specific high severity scenarios, requiring evidence of an
• 0:58
  Speaker 1: imminent risk capable of causing over $ 1 billion in financial
• 1:02
  Speaker 1: damage or resulting in more than 50 fatalities or severe injuries. Furthermore,
• 1:09
  Speaker 1: these intervention thresholds are narrow in scope, limiting emergency government
• 1:14
  Speaker 1: actions exclusively to threats involving biological weapons, widespread cyberattacks, or
• 1:20
  Speaker 1: instances where an advanced software system evades developer control. Under
• 1:25
  Speaker 1: the proposed enforcement architecture, the Department of Commerce would serve
• 1:29
  Speaker 1: as the sole federal authority permitted to petition a judge
• 1:32
  Speaker 1: in the United States District Court for the District of
• 1:35
  Speaker 1: Columbia to suspend a model's deployment, while explicitly barring all
• 1:40
  Speaker 1: other executive departments and regulatory bodies from taking independent administrative action.
• 1:46
  Speaker 1: To maintain ongoing oversight, the legislative draft mandates that artificial
• 1:52
  Speaker 1: intelligence developers undergo periodic evaluations conducted by independent auditors registered
• 1:58
  Speaker 1: with the Department of Commerce. These certified auditing entities would
• 2:02
  Speaker 1: be required to notify federal authorities within 72 hours if
• 2:07
  Speaker 1: testing reveals an imminent risk meeting the statutory threshold. Representatives
• 2:12
  Speaker 1: from Anthropic confirmed that the proposal reflects principles from the
• 2:16
  Speaker 1: firm's previously published policy framework, describing the measures as sensible
• 2:22
  Speaker 1: national safeguards designed to manage catastrophic risks while providing regulatory
• 2:27
  Speaker 1: certainty for technological innovation across the domestic market.