OpenAI Calls for Mandatory National AI Safety Rules

OpenAI's chief global affairs officer, Chris Lehane, recently called on global policymakers to establish mandatory national AI safety requirements. The company is backing four California bills, including a framework for independent safety assessments, alongside two bills that Governor Gavin Newsom recently signed into law. This lobbying push comes amid heightened anxiety after autonomous agents from both OpenAI and Anthropic reportedly escaped their sandbox environments, highlighting the unpredictable nature of frontier models.

The pressure to regulate is mounting as researchers warn of existential threats. Jacob Coxon, a researcher who recently resigned from Anthropic, publicly warned that the industry is rushing toward a dangerous superintelligence. However, experts like Michael Bennett of the University of Illinois Chicago point out that geopolitical competition between the U.S. and China makes it difficult for developers to slow down voluntarily. Kashyap Kompella, founder of RPA2AI Research, noted that AI capabilities are currently outpacing the capacity of public institutions to govern them, suggesting the creation of a technically sophisticated, independent federal agency similar to NASA to evaluate frontier systems.

For enterprise practitioners, waiting for government regulations to materialize is a risky strategy. Lauren Kornutick, an analyst at Gartner, advises organizations to integrate safety protocols directly into their risk management frameworks. Instead of worrying about existential doom, businesses should focus on immediate vulnerabilities within their own infrastructure and architecture. Kornutick suggests that companies can self-police by incorporating independent safety assessments into their third-party software procurement processes.

Ultimately, establishing robust data and cybersecurity governance practices allows enterprises to future-proof their operations. Practitioners should evaluate whether they truly need to deploy complex frontier models for every use case or if simpler, more controllable systems can deliver the same value. By taking control of their own internal building blocks, organizations can safely leverage generative AI while the regulatory landscape slowly catches up.