Updated Sep 2, 2026
/Trump and Rubio seal Venezuela oil deal covering 65 billion barrels across 17 fields/Bond rout gathers pace as borrowing costs hit multi-decade highs/Van Jones calls El-Sayed's total Israel arms embargo 'insane,' cites Obama legacy on defense/Memecoin buys on Robinhood Wallet and Fomo coded as "digital media," sidestepping card-network crypto rules/Caring Brands (CABR) Closes $4.6 Million Initial Series B Preferred Stock Private Placement/Sanmar Group takes majority stake in AltEons Energy to back 1.5GW India pipeline/Trump and Rubio seal Venezuela oil deal covering 65 billion barrels across 17 fields/Bond rout gathers pace as borrowing costs hit multi-decade highs/Van Jones calls El-Sayed's total Israel arms embargo 'insane,' cites Obama legacy on defense/Memecoin buys on Robinhood Wallet and Fomo coded as "digital media," sidestepping card-network crypto rules/Caring Brands (CABR) Closes $4.6 Million Initial Series B Preferred Stock Private Placement/Sanmar Group takes majority stake in AltEons Energy to back 1.5GW India pipeline

Trump administration tells AI companies no safety testing required for open-weight models

WASHINGTON, Aug. 5. The Trump administration has informed artificial intelligence companies that it will not require safety testing for open-weight AI models, according to sources familiar with the discussions. The communication addresses a specific class of AI release where model weights are publicly distributed, allowing any party to download, run, or modify the underlying system independent of the original developer.

By Julian Merrick2 min read
Share

Key takeaways

  • The Trump administration has told AI companies it will not require safety testing for open-weight AI models, according to sources familiar with the discussions.
  • The decision applies specifically to open-weight models, where model weights are publicly distributed so anyone can download, run, or modify the system independent of the original developer.
  • As described, the position would leave open-weight model developers without a federal safety-testing obligation ahead of a public release.
  • Sources did not identify which companies were notified or how the communication was delivered.
  • Unlike proprietary API-based systems that can be monitored, restricted, or taken offline, a developer that releases weights publicly has no comparable control once distribution occurs.

WASHINGTON, Aug. 5. The Trump administration has informed artificial intelligence companies that it will not require safety testing for open-weight AI models, according to sources familiar with the discussions. The communication addresses a specific class of AI release where model weights are publicly distributed, allowing any party to download, run, or modify the underlying system independent of the original developer.

The administration's position

The sources, who described the discussions, did not identify which companies were notified or in what form the communication was delivered. The position, as described, would leave open-weight model developers without a federal safety-testing obligation ahead of a public release.

Open-weight models and the scope of the decision

Open-weight releases are structurally different from proprietary systems. A company that controls access through an API can monitor use, restrict accounts, or take a model offline. A developer that releases weights publicly has no comparable mechanism once distribution occurs. The administration's reported decision applies to that second category of release.

Frequently asked

What did the Trump administration tell AI companies?

It informed AI companies that it will not require safety testing for open-weight AI models, according to sources familiar with the discussions.

What is an open-weight AI model?

It is a class of AI release in which model weights are publicly distributed, allowing any party to download, run, or modify the underlying system independent of the original developer.

Why does the distinction between open-weight and proprietary models matter here?

A company controlling access through an API can monitor use, restrict accounts, or take a model offline, but a developer that releases weights publicly has no comparable mechanism once distribution occurs, and the decision applies to that second category.

Which companies were notified of the decision?

The sources did not identify which companies were notified or in what form the communication was delivered.

Does the decision apply to proprietary API-based models?

No; the administration's reported decision applies to open-weight releases where weights are distributed publicly, not to proprietary systems controlled through an API.