The Donald Trump administration is weighing a significant shift in its artificial intelligence strategy.
With plans that could place the Pentagon at the centre of testing AI systems before they are cleared for use across government. According to a report by Axios, the proposal would require advanced AI models to undergo safety and risk evaluations led by the Pentagon before being deployed across federal, state, and local agencies. The move signals a more structured — and potentially softer — approach compared to the administration’s earlier hardline stance on AI oversight.
Pentagon’s expanded role
Under the proposed framework, the Pentagon would act as a key gatekeeper, assessing whether emerging AI systems meet security and reliability standards before being integrated into public sector operations.
Discussions are being driven by the White House’s Office of the National Cyber Director, which recently held consultations with tech firms, cybersecurity experts, and industry groups to evaluate the risks posed by increasingly powerful AI tools.
Rising concerns over advanced AI
A major trigger for these talks has been the emergence of frontier AI models such as Mythos Preview developed by Anthropic. The system has drawn attention for its advanced capabilities, particularly in identifying and potentially exploiting cybersecurity vulnerabilities.
Officials are said to be exploring a broader security framework that would not only test AI models but also address risks like “model theft” and misuse by adversarial actors, including concerns linked to China.
Policy shift still in flux
Despite the ongoing discussions, no final decision has been taken. A White House official emphasised that any formal policy direction would come directly from the president, with current deliberations still at an exploratory stage.
The evolving stance reflects a reassessment of earlier decisions, including a directive that restricted government agencies from working with Anthropic. At the time, the Pentagon had classified the company as a supply-chain risk amid disagreements over safeguards for military AI use.
Industry pushback and next steps
Anthropic has challenged that designation and is currently engaged in legal action against the Department of Defense. The company has also clarified that its most advanced system will not be publicly released, opting instead for controlled testing through its “Project Glasswing” initiative.
As AI capabilities continue to accelerate, the administration’s approach appears to be shifting from outright restriction toward controlled adoption — with the Pentagon potentially playing a central role in determining how, and how safely, these technologies are deployed.
Comments are closed.