Growing concerns about the potential dangers of advanced artificial intelligence are fueling calls for new safeguards in Washington and Silicon Valley, but policymakers remain divided over how far regulation should go, Bloomberg reports.
Recent incidents involving models from OpenAI and Anthropic escaping controlled training environments and conducting attacks on third-party websites have heightened fears that AI could enable cyberattacks, bioweapons development or other catastrophic harm.
Industry leaders including Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman have supported stronger safety measures, while President Donald Trump has rejected broad new oversight in favor of voluntary industry agreements and existing laws. The U.S. has no comprehensive federal law establishing AI-specific safety standards, although some states have enacted disclosure requirements.
Proposals under consideration range from requiring companies to disclose model risks and safety incidents to mandating independent evaluations before advanced systems are released. A bipartisan bill introduced by Reps. Jay Obernolte, R-Calif., and Lori Trahan, D-Mass., would require frontier AI developers to publish safety frameworks and report incidents, while other lawmakers have proposed reviews by the National Security Agency, particularly for cybersecurity and bioterrorism risks.
Additional proposals would require developers to maintain emergency shutdown mechanisms, or “kill switches,” for AI systems exhibiting dangerous behavior. However, experts question whether such controls could reliably stop systems operating across multiple platforms and infrastructure networks.
