OpenAI U-Turn: Company Urges Congress to Pass Mandatory Capability-Based AI Safety Law

When the Atomic Energy Act of 1946 placed nuclear R&D under federal control, the initiative was led by Manhattan-project scientists who recognised their creation’s risks; OpenAI’s plea echoes that self-regulation-turned-regulation moment. Over the past two decades, dominant tech platforms have repeatedly tried to shape rules— from Facebook’s 2018 call for internet privacy standards to Microsoft’s 2021 lobbying for cybersecurity mandates—by proposing regimes they already meet and smaller rivals cannot. OpenAI’s capability-triggered model would similarly entrench frontier labs at the top while answering growing anxiety over ‘runaway’ agents. Whether Congress acts or not, the announcement signals an inflection where AI developers publicly concede that voluntary guard-rails are insufficient, nudging the U.S. toward a licensing era reminiscent of the 19th-century Interstate Commerce Act that reined in railroads. On a 100-year horizon, the episode may mark the moment society began demanding legally enforceable brakes on self-improving machines—potentially as consequential for governance as early aviation treaties were for the skies— or it may be remembered as corporate positioning that delayed deeper, democratically driven constraints. Time will tell which narrative sticks.