Trump administration exempts open-weight AI models from voluntary safety tests

The Executive Order and Voluntary Framework

On June 2, 2026, President Trump signed an executive order titled "Promoting Advanced AI Innovation and Security," which calls on federal agencies to obtain voluntary agreements with top AI developers for pre-release cybersecurity evaluations of their most advanced models . The framework explicitly bans mandatory licensing or any new regulatory bodies .

Rather than building a new regulatory agency, the order directs the Treasury, Defense, Commerce, and Homeland Security departments to secure agreements with AI developers for model testing . Developers who opt into the voluntary framework should plan for a 30-day pre-release government access window .

The Open-Weight Exemption

In an August 4, 2026 meeting with AI developers at the White House, Trump advisers confirmed that open-weight models — AI systems with publicly accessible core components that allow users to download, modify, and run the technology on their own infrastructure — will not be put through the voluntary safety tests . The White House also told companies it would exempt Chinese open-weight models from government testing under the framework .

Representatives from Meta, Anthropic, Google, Nvidia, and OpenAI attended the meeting, according to five sources familiar with the discussions . The meeting followed disclosures from OpenAI and Anthropic that their AI tools had breached the systems of other companies, raising concerns among lawmakers about whether increasingly capable AI models could be used to conduct or facilitate cyberattacks .

Scope of Testing: Cybersecurity Focus

The finalized tests focus on measuring the hacking and cybersecurity capabilities of frontier AI models, not broader safety or bias evaluations . The designation thresholds for what counts as a "covered frontier model" will be set through a classified NSA process, giving developers little visibility into where the line falls .

Concerns Raised by the Exemption

Safety Gap

Critics argue that exempting open-weight models creates a significant blind spot. Open-weight models can be freely downloaded, modified, and weaponized by bad actors, and they lack the guardrails that proprietary labs build into their closed systems . Anthropic CEO Dario Amodei has argued that open-weight models are harder to keep safe because, once released, the developer loses control over how the model is used .

Competitive Distortion

Because the framework only scrutinizes leading proprietary models from US labs like OpenAI and Anthropic — while leaving open-weight models, including Chinese ones, untouched — some worry US frontier labs will face a regulatory disadvantage against open-source competitors abroad .

Definitional Ambiguity

The classified NSA process for setting designation thresholds means developers have little visibility into where the line between covered and uncovered models falls, creating uncertainty about which models might eventually face scrutiny .

Industry Reactions

Closed-Model Labs: OpenAI and Anthropic

OpenAI and Anthropic have been warning policymakers about the risks of powerful open-weight models, positioning themselves for greater scrutiny of open-source systems . Critics, including Trump administration adviser David Sacks, say these labs are acting out of commercial self-interest, and that scrutiny could amount to regulatory capture: rules intended to improve AI safety could instead entrench the largest companies by making it harder for competitors to release models .

Open-Weight Advocates: Meta and Nvidia

Meta and Nvidia benefit directly from the exemption, as their Llama and Nemotron models remain outside the testing framework .

Startups and the "Little Tech" Coalition

Nearly 200 companies, including Proton and Y-Combinator, urged the administration not to restrict access to Chinese open-weight AI models, arguing that doing so could cripple the next generation of US startups . The newly-formed Little Tech Association sent letters to President Trump and Commerce Secretary Howard Lutnick making this appeal .

Silicon Valley Whiplash

The New York Times characterized the administration's approach as creating whiplash — rolling back Biden-era AI safety mandates, then introducing a voluntary framework, then exempting open-weight models — leaving the industry uncertain about what rules, if any, will apply . Saif Khan, a former adviser on emerging technology in the Biden administration, called the new White House approach "opaque, almost vibes-based" and potentially more damaging to the AI industry than traditional regulation .