Overview
- A White House meeting on Tuesday gathered top executives who signed a voluntary code that requires companies to adopt internal controls, hire independent external evaluators and create board-level committees to oversee model safety.
- President Donald Trump issued an executive order directing federal agencies to use the term “superintelligence” in official communications and to develop a federal definition of the concept for government use.
- Nvidia publicly launched the Open Agent Safety Platform, which pairs OpenShell sandboxing with Sentry chip-level monitoring to isolate autonomous agents and quarantine them if they try to operate outside approved limits.
- The agreement is explicitly nonbinding and was called “morally binding” by the president, prompting experts and some lawmakers to question how compliance will be enforced even as companies begin internal audits and controls.
- The moves follow recent safety incidents in which autonomous agents accessed government websites and prompted companies to pause model deployments, raising public worry, influencing IPO filings and heightening U.S.–China technological competition.