Anthropic is suing the Pentagon and briefing the White House at the same time, which tells you everything about how AI companies are learning to play the power game.
The Summary
- Anthropic co-founder Jack Clark confirmed the company briefed the Trump administration on Mythos, their latest model with what they claim are powerful cybersecurity capabilities.
- The company simultaneously hired Ballard Partners, a Trump-linked lobbying firm, while actively fighting the Pentagon in court.
- This is the new playbook: sue one branch of government while courting another, all while building the models that could define national security infrastructure.
The Signal
Jack Clark explained the dual strategy at the Semafor World Economy summit this week. The company is actively litigating against the Pentagon while maintaining direct lines to the executive branch. That's not contradiction. That's strategy.
The Mythos model sits at the center of this. Anthropic claims it has significant cybersecurity capabilities, the kind that make it simultaneously valuable to brief the White House about and dangerous to let the Defense Department control without oversight. The company is threading a needle: we'll help you understand what we're building, but we won't let you dictate how we build it.
"Anthropic is fighting the Pentagon while briefing the President, which is either very smart or very reckless depending on how the next six months play out."
The Ballard Partners hiring reveals the infrastructure behind this strategy. You don't hire a Trump-connected lobbying firm for general relationship building. You hire them when you need specific doors opened, specific calls returned, specific meetings that happen before decisions get made. Anthropic is building political capital while their legal team files motions.
This matters because it maps the future of AI governance. The old model was: build the thing, then deal with regulators. The new model is: build the thing, sue the regulators who overstep, brief the politicians who fund them, and hire the lobbyists who know which conversations happen off the record. Anthropic is running this playbook in real time.
The cybersecurity angle makes this more urgent. Models with serious offensive or defensive cyber capabilities aren't just products. They're weapons, infrastructure, and intelligence assets rolled into one. The government wants access. The companies want autonomy. The lawyers get rich. And somewhere in the middle, the actual policy gets written by whoever shows up to the meetings.
The Implication
Watch what happens when other frontier AI labs copy this strategy. If briefing the White House while suing the Pentagon works for Anthropic, OpenAI and Google won't be far behind. The next 18 months will define whether AI companies can maintain independence from government control or whether models like Mythos end up effectively nationalized through procurement contracts and security clearances.
For anyone building in this space, the lesson is clear: your technical roadmap and your government relations strategy are the same thing now. You can't build frontier models without a plan for who gets briefed, who gets sued, and who gets hired to make the phone ring.