U.S. government proposes deploying Anthropic AI model amid safety concerns
The U.S. Office of Management and Budget announced plans to deploy Anthropic's Mythos AI model across major federal agencies despite prior designation of Anthropic as a supply chain threat due to AI safety disputes. Anthropic had restricted Mythos access because internal testing showed it could identify critical software vulnerabilities, creating risk that malicious actors could exploit it for cyberattacks.
What happened
The U.S. Office of Management and Budget announced on April 16, 2026, that it was developing safeguards to enable federal agencies to use Anthropic's Mythos AI model. OMB federal Chief Information Officer Gregory Barbacia sent an email to officials at the Defense Department, Treasury Department, Commerce Department, Department of Homeland Security, Justice Department, and State Department stating that protective measures were being prepared, with additional information to be provided within weeks.
Anthropic had restricted Mythos access due to safety concerns. Internal testing showed the model could identify critical software bugs that only world-class hackers typically find, and Anthropic expressed concern that malicious actors could exploit it to steal data or destroy networks. One source told Bloomberg that giving Mythos to ordinary hackers would be equivalent to transforming ordinary soldiers into special forces operatives.
Reported impact
The proposed deployment came amid ongoing conflict between Anthropic and the Trump administration. The U.S. Department of Defense had previously designated Anthropic as a supply chain threat due to disputes over AI safety measures, and Anthropic had blocked the government's technology ban through a court order the previous month.
Index analysis
Anthropic had briefed government officials including the Cybersecurity and Infrastructure Security Agency and the Center for AI Standards and Innovation on Mythos's full capabilities including offensive and defensive cybersecurity applications before the model's launch.
On April 16, 2026, Anthropic released Claude Opus 4.7, which maintained lower cybersecurity capabilities than the Mythos Preview. The model incorporated automated detection and blocking of requests for prohibited or high-risk cybersecurity uses. Anthropic stated it planned to apply lessons learned through this model toward broader deployment of Mythos-level models.
Governance controls
Anthropic employed safety measures in Claude Opus 4.7 to automatically detect and block requests for cybersecurity applications classified as prohibited or high-risk.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.