OpenAI publicly called for mandatory, capability-based national AI safety regulation in the United States after previously warning that heavy rules could let China gain leadership. Chris Lehane, OpenAI’s chief global affairs officer, wrote in a September 9 blog post titled “The AI policy window is open. We need to act.” that the US needs a durable federal framework for frontier AI, including common testing, independent assessments, cybersecurity protections and clear incident-reporting requirements for the “handful of well-resourced laboratories” building the most capable systems.

Lehane said OpenAI is now supporting some bills it had not endorsed before “after reconsidering in light of the recent jump in capabilities we have seen.” The article describes that line as the company’s closest admission that its own systems may have crossed a risk threshold it did not expect. OpenAI’s proposed oversight agenda includes support for four California bills headed to Governor Gavin Newsom’s desk.

The California bills would set up a process for designating qualified independent organizations to assess AI risks, create registration and accountability requirements for AI auditors, require age assurance, risk assessments and safeguards for children and teens using companion chatbots, and strengthen screening standards for gene-synthesis providers to curb AI-enabled biological threats. OpenAI also endorsed mandatory federal safety rules for frontier labs, independent technical assessments, stronger cybersecurity and incident reporting.

OpenAI additionally backed industry-led monitoring standards for misalignment (AI behaving against intended goals) and serious incidents, starting voluntarily but meant to shape federal policy. Lehane also called for international coordination on capability measurement, human control and decisions on when to slow or stop development, “even if that means slowing the advancement of model capabilities.”

The policy shift came days after OpenAI chief scientist Jakub Pachocki warned that no lab has solved AI alignment at current scales and that voluntary slowdowns may soon be necessary. The article also cites the resignation of Jacob Coxon, 27, a former Anthropic and OpenAI researcher who said leading labs are racing toward “self-improving superintelligence” and warned AI systems could be “out of control” by the end of 2027.

The article says GPT Astra’s launch added to safety concerns after Nvidia CEO Jensen Huang described it as Artificial General Intelligence. Reuters reported that the latest version of OpenAI’s large language model reached the highest safety threshold for autonomous cyberattacks and could hide its internal reasoning. Lehane’s position contrasts with G20 countries’ light-touch AI governance framework backed by the US and China.

OpenAI publicly called for mandatory, capability-based national AI safety regulation in the United States after previously warning that heavy rules could let China gain leadership. Chris Lehane, OpenAI’s chief global affairs officer, wrote in a September 9 blog post titled “The AI policy window is open. We need to act.” that the US needs a durable federal framework for frontier AI, including common testing, independent assessments, cybersecurity protections and clear incident-reporting requirements for the “handful of well-resourced laboratories” building the most capable systems.

Lehane said OpenAI is now supporting some bills it had not endorsed before “after reconsidering in light of the recent jump in capabilities we have seen.” The article describes that line as the company’s closest admission that its own systems may have crossed a risk threshold it did not expect. OpenAI’s proposed oversight agenda includes support for four California bills headed to Governor Gavin Newsom’s desk.

The California bills would set up a process for designating qualified independent organizations to assess AI risks, create registration and accountability requirements for AI auditors, require age assurance, risk assessments and safeguards for children and teens using companion chatbots, and strengthen screening standards for gene-synthesis providers to curb AI-enabled biological threats. OpenAI also endorsed mandatory federal safety rules for frontier labs, independent technical assessments, stronger cybersecurity and incident reporting.

OpenAI additionally backed industry-led monitoring standards for misalignment (AI behaving against intended goals) and serious incidents, starting voluntarily but meant to shape federal policy. Lehane also called for international coordination on capability measurement, human control and decisions on when to slow or stop development, “even if that means slowing the advancement of model capabilities.”

The policy shift came days after OpenAI chief scientist Jakub Pachocki warned that no lab has solved AI alignment at current scales and that voluntary slowdowns may soon be necessary. The article also cites the resignation of Jacob Coxon, 27, a former Anthropic and OpenAI researcher who said leading labs are racing toward “self-improving superintelligence” and warned AI systems could be “out of control” by the end of 2027.

The article says GPT Astra’s launch added to safety concerns after Nvidia CEO Jensen Huang described it as Artificial General Intelligence. Reuters reported that the latest version of OpenAI’s large language model reached the highest safety threshold for autonomous cyberattacks and could hide its internal reasoning. Lehane’s position contrasts with G20 countries’ light-touch AI governance framework backed by the US and China.