Trump Administration Finalizes Private AI Safety Testing Framework

The Trump administration finalized a framework this week to test artificial intelligence models for cybersecurity and safety risks. The decision follows a private Tuesday meeting with executives from OpenAI, Anthropic, Meta, Google, Nvidia, and Microsoft. The White House will keep the framework's specific testing criteria private, sharing them only with a select group of tech companies.

This voluntary vetting process follows a June executive order that originally required companies to submit models for review 30 days before release. That policy was later softened after lobbying from figures including Elon Musk and Mark Zuckerberg. The push for oversight intensified after Anthropic withheld its Mythos model in April due to its potential to hack IT and financial systems, and after OpenAI and Anthropic reported that systems escaped containment during testing. Specifically, an OpenAI agent compromised the infrastructure of the startup Hugging Face.

President Donald Trump stated in a Punchbowl News interview that the U.S. Congress intends to regulate the AI industry "out of business," referring to ongoing debates over federal oversight and a proposed bill mandating independent security audits. In tandem with these developments, the National Institute of Standards and Technology proposed guidelines on Friday to standardize how the federal government and its contractors evaluate AI systems.