U.S. President Donald Trump has opened the door to a potential reset with Anthropic, indicating the firm could regain standing with the United States Department of Defense after a high-profile rift over how advanced artificial intelligence should be deployed in national security settings.
Speaking on CNBC, Trump said the company was “shaping up” following recent White House talks, adding that a deal with the Pentagon was “possible.”
“They ?came to the White House a few days ago, and we had some very good ?talks with them,” Trump said. “And I think they’re shaping up. They’re very smart, and I think ?they can be of great use. I like smart people … I think we’ll get along with them just fine.”
The comments mark a notable shift from February, when the administration directed federal agencies to cease engagement with Anthropic, triggering a Pentagon designation that labeled the firm a supply-chain risk.
That designation effectively sidelined Anthropic from defense work and barred its tools from use by military personnel and contractors after a six-month wind-down period. The decision reflected deeper concerns within the Pentagon about operational control, reliability, and the governance frameworks surrounding frontier AI systems.
A breakdown over guardrails was at the core of the dispute. Anthropic had sought binding assurances that its models would not be used for domestic surveillance or autonomous weapons, conditions that ran counter to the Pentagon’s push to integrate AI across intelligence gathering, cyber operations, and battlefield decision-support systems. Defense officials, for their part, viewed such restrictions as incompatible with evolving military doctrine, particularly as rivals accelerate their own AI capabilities.
The standoff quickly escalated into a broader confrontation over the role of private AI labs in national security. Anthropic argued that unchecked deployment risked misuse and unintended escalation, while policymakers framed access to leading-edge models as essential to maintaining advantage. The result was a rare public rupture between a frontier AI developer and the U.S. defense establishment.
Anthropic’s legal challenge in March underscored the stakes. The company moved to overturn the blacklist, arguing that the designation was both procedurally flawed and strategically counterproductive. Court rulings have so far been mixed, with a federal appeals court allowing the restrictions to remain in place for now, even as parallel litigation exposed divisions within the judiciary over how such cases should be handled.
The tone, however, has begun to shift—and the catalyst appears to be technological as much as political.
Anthropic’s unveiling of its Mythos model is believed to have reframed the debate. The system is widely described as a step change in capability, particularly in cybersecurity contexts, where it can identify vulnerabilities and simulate exploitation pathways with a level of sophistication that raises both defensive and offensive implications. Rather than releasing the model broadly, Anthropic opted for a controlled rollout under its “Project Glasswing” programme, inviting select partners, including JPMorgan Chase and cybersecurity firms, to stress-test the system and develop safeguards.
This approach appears to have resonated in Washington. By restricting access and emphasizing defensive use cases, Anthropic has sought to demonstrate that frontier AI can be deployed responsibly without ceding ground. For policymakers, the model’s capabilities also highlight a reality that complicates outright exclusion: tools of this sophistication are increasingly integral to cybersecurity, an area where the Pentagon faces constant and evolving threats.
Anthropic Chief Executive Dario Amodei met with senior administration officials last week in what both sides described as constructive discussions. The company said talks centered on shared priorities, including securing U.S. leadership in AI, strengthening cyber defenses, and establishing safety frameworks for advanced systems.
Trump’s latest remarks suggest a reconsideration by the Pentagon. His emphasis on “smart people” and potential collaboration points to a recognition that sidelining a leading AI developer comes with costs, particularly as global competition intensifies. However, his characterization of the company as aligned with “the radical left” indicates that political considerations remain intertwined with policy decisions.
The Pentagon’s initial ban was not absolute. Exemptions tied to national security needs allowed for limited engagement, signaling that even at the height of the dispute, there was acknowledgment of Anthropic’s technical value. What appears to be emerging now is a pathway toward conditional reintegration—one that would likely involve stricter oversight, clearer usage boundaries, and closer coordination with government stakeholders.
The broader context is a shifting regulatory landscape. Governments are still grappling with how to harness increasingly autonomous and capable AI systems without losing control over their application. The Anthropic case has become a test scenario, illustrating the tension between innovation and constraint, and raising questions about how much influence private developers should have over the use of their technologies in state functions.
Mythos has intensified that conversation. Its capabilities blur the line between defensive cybersecurity tools and offensive cyber potential, making governance more complex. U.S. government officials issued a warning to the banks executives, following the release of Mythos.
However, many see the White House’s current move as less a simple reconciliation than a renegotiation of terms. The initial rupture exposed a fundamental mismatch between the incentives of Silicon Valley and the imperatives of national security. The current shift suggests both sides are moving toward a more transactional arrangement.
