On June 5, Dario Amodei, the CEO of Anthropic, published a guest essay in The New York Times urging U.S. lawmakers to formulate transparency regulations for artificial intelligence companies instead of implementing the decade – long freeze on state – level regulations proposed in President Donald Trump’s technology bill.
Amodei shared an internal evaluation by Anthropic, where its latest model threatened to expose a user’s private emails if the operator didn’t cancel a shutdown plan. He also mentioned separate examinations that found OpenAI’s GPT – 3 model drafting self – preservation code and Google’s Gemini model nearing the ability to assist in cyberattacks. Amodei compared these exercises to wind – tunnel tests for aircraft, aiming to identify flaws before public release. He recognized the documented productivity improvements in drug development and medical triage but emphasized that safety teams must detect and prevent risks “before they find us”.
Amodei pointed out that Anthropic, OpenAI, and Google DeepMind currently publish responsible scaling policies and give independent researchers early access to advanced systems, despite lacking a legal obligation to do so. He noted that there is currently no federal law mandating such disclosures.
The Senate draft bill would prohibit states from passing AI – related laws for ten years to avoid a fragmented regulatory landscape. Amodei believes this timeline is too long compared to the pace of technological progress and could leave regulators without effective oversight. He asked Congress and the White House to set a unified requirement, compelling developers of the most powerful models to publish their testing methods, risk – mitigation measures, and release criteria on their websites. Anthropic already discloses these details under its responsible scaling policy, and Amodei argued that codifying similar practices across the industry would allow the public and legislators to monitor capability advancements and decide if further action is needed.
In addition, Amodei supported export controls on advanced chips and the military’s adoption of trusted systems to counter China. He also suggested that states could adopt limited disclosure rules that would give way to a future federal framework. Once Congress enacts a national standard, a supremacy clause could supersede state measures, maintaining regulatory uniformity without impeding short – term local actions.
Related topic:
- The US SEC has ended its multi-year review of Coinbase’s financial disclosures without requiring the company to revise or restate them
- Trump criticized Powell for playing politics and claimed that the president had the right to dismiss the chairperson of the Federal Reserve
- Sumitomo Mitsui Securities has made a strategic investment in a blockchain fund to facilitate the integration of Japanese finance with Web3
