The Trump administration said on June 2 that it would seek voluntary agreements to let federal agencies test advanced AI models for cybersecurity risks for up to 30 days before release outside the government.

The move expands Washington’s role even though President Donald Trump had repealed Joe Biden’s 2023 AI order requiring developers to share safety test results with the government before release.

The executive order directed the Treasury, Defense, Commerce, and Homeland Security departments, along with other agencies, to secure the agreements and strengthen federal cyber defenses.

It also asked Treasury Secretary Scott Bessent to work with AI developers and critical infrastructure providers, such as banks and hospitals, to identify software vulnerabilities and help develop patches.

One focus is Mythos, a model from Anthropic, a US-based AI startup, which the company revealed in April as Mythos Preview and said showed unusually strong cybersecurity capabilities in testing.

The move also builds on a Commerce Department process.

The National Institute of Standards and Technology’s Center for Artificial Intelligence Standards and Innovation is already conducting testing on models from Google DeepMind, Microsoft, and xAI in classified environments under renegotiated voluntary agreements.

Like earlier White House AI safety pledges, the arrangements would be voluntary and have no explicit enforcement mechanism.

🔗 Source: Reuters