OpenAI released a set of safety proposals on Monday calling for strict international standards and focused research into the risks of autonomous artificial intelligence (AI).
The move comes as high-profile resignations, corporate proposals, and government interventions signal a critical turning point for the AI industry, where developers face growing pressure to manage risks surrounding foundation models and automated systems.
OpenAI’s proposal focuses heavily on recursive self-improvement (RSI), a computing technique that allows AI foundation models to upgrade and refine their own capabilities without human intervention.
While developers view RSI as a breakthrough path forward, safety experts warn it could lead to catastrophic loss of human control over complex digital infrastructure.
“Navigating this transition safely requires alignment research to keep pace with these capabilities so that the systems we and others build remain aligned with human values,” OpenAI shared in a blog post, cautioning that fully autonomous RSI should not be pursued until robust safeguards are established. (OpenAI CEO Sam Altman is scheduled to address the UN Security Council on Wednesday.)
The company called for international cooperation built on existing AI safety institutes to create standardized oversight, referencing recent security breaches as a preview of potential systemic vulnerabilities.
OpenAI’s announcement follows a turbulent period for the industry.
Former Anthropic and OpenAI researcher Jacob Coxon sparked global debate after resigning, alleging that top AI firms were “gambling with our lives.”
Shortly after, Anthropic CEO Dario Amodei advocated for slowing the pace of model development and embedding third-party auditors within AI companies, a concept endorsed by Altman and Tesla Inc. CEO Elon Musk. Consequently, a coalition of independent evaluators is pushing for minimum access guarantees and anti-retribution protections to audit cutting-edge models effectively.
As AI developers debate self-regulation, government officials are defining the boundaries of legal and geopolitical responsibility. Speaking on CNBC’s Squawk Box on Monday, Treasury Secretary Scott Bessent rejected calls for federal liability protections for tech firms, stating firmly, “It is humans who are responsible, not the AI.”
The stance reflects a delicate balance within the Trump administration, which favors rapid AI expansion and data center infrastructure while maintaining that corporate leadership must shoulder legal accountability.
Bessent’s comments followed a marathon 12-hour bilateral meeting on Sunday with Chinese Vice Premier He Lifeng ahead of an upcoming summit between President Donald Trump and Chinese President Xi Jinping in Washington.
During the talks, U.S. and Chinese officials agreed to formalize a dedicated channel for future AI-related security incidents, aiming to align the global superpowers on mutual threats such as rogue autonomous agents, state-sponsored cyber risks, and bioweapon proliferation. Further discussions are slated for the Asia-Pacific Economic Cooperation summit in Shenzhen later this year.
The high-level AI discussions occurred alongside pressing economic and foreign policy negotiations. Bessent noted that talks with Beijing focused heavily on the Nov. 10 expiration date of the U.S.-China trade truce, as well as broader sanctions targeting financial entities supporting Iran.
Addressing recent domestic economic friction, Bessent commented on the Federal Reserve’s unanimous vote last week to raise benchmark interest rates to a target range of 3.75% to 4.00% to combat inflation.
The Treasury Secretary expressed optimism that interest rates and energy prices would stabilize once active Middle Eastern conflicts subside and global oil markets rebalance.
