Anthropic reconciles with Trump: new Claude Mythos model for cybersecurity

Anthropic may be regaining the Trump administration's favor — and the key to this lies in a new specialized model, Claude Mythos Preview, designed for cybersecurity tasks. Over the past two months, relations between Anthropic and the White House have been extremely tense. The administration publicly called the company a "left-radical woke organization" populated with "left-wing cranks" and characterized it as a threat to national security.

A sharp escalation occurred in late February, when Anthropic refused to compromise on two key issues: the company would not agree to provide technologies for mass domestic surveillance or for creating fully autonomous lethal weapons without human involvement in the decision-making chain. This stance cost Anthropic its relationship with the Pentagon — a major consumer of AI solutions. Prior to this, the company's technologies were actively used across the government sector, and the breach hurt both sides.

Anthropic lost lucrative contracts, while agencies lost access to one of the most powerful commercial AI stacks. However, Claude Mythos Preview changes the picture. The model was developed under the internal codename Project Glasswing and is specifically oriented toward cybersecurity tasks: threat analysis, vulnerability detection, incident response.

This is precisely the class of applications that does not contradict Anthropic's ethical constraints, yet directly serves the interests of the military and intelligence services. According to sources close to negotiations, the emergence of Mythos opens the possibility of restarting dialogue between Anthropic and the U.S.

government. The company is demonstrating willingness to collaborate where it does not violate its red lines — and cybersecurity turns out to be precisely that point of intersecting interests. The story is noteworthy because it demonstrates: even amid political turbulence, pragmatism prevails over rhetoric.

The administration, which publicly attacked the company, apparently stands ready to return to the negotiating table once a product addressing actual government needs appears. For Anthropic, this is a signal that adherence to ethical principles and work with the government are not mutually exclusive — if the right point of effort is chosen correctly.