Analysis
Commerce Secretary Howard Lutnick told Axios that the Trump administration now trusts Anthropic, months after the two clashed publicly over guardrails on the Pentagon's use of the company's AI models, Axios reported, a reconciliation Forbes also covered. Asked directly whether he trusts Anthropic CEO Dario Amodei, Lutnick said: "We trust Anthropic. They've done what we asked. They're back on the right side. So the answer is: Yes."
What the Original Fight Was About
The dispute wasn't abstract policy disagreement -- Anthropic insisted on specific guardrails restricting how the Pentagon could deploy its models, including prohibiting the Department of Defense from using Claude for mass surveillance of Americans or for autonomous weapons systems. That put the company at odds with a Defense Department that wanted fewer restrictions on how it could procure and deploy frontier AI capability, at a moment when the administration was broadly pushing a deregulatory posture on AI development -- the same posture reflected in the Carolina Principles Pulse covered from this week's G20 summit. Anthropic taking a restrictive position on military use while the administration pushed hands-off AI policy elsewhere created a specific, visible friction point between one of the two leading US frontier labs and the government that regulates and, increasingly, procures from it.
What Actually Changed
Lutnick's phrasing -- "they've done what we asked" -- implies Anthropic adjusted its position rather than the administration softening its demands, though neither Lutnick nor Anthropic detailed exactly what changed in the guardrails themselves. The more concrete evidence of a genuine thaw is personnel: Anthropic co-founder Tom Brown has taken a more prominent role representing the company to the White House, including headlining at this week's G20 Innovation Ministerial alongside Commerce and OSTP officials. Elevating a co-founder into the DC relationship, rather than leaving it to policy staff, is the kind of investment a company makes when it's decided the relationship needs direct, senior attention rather than routine government-affairs management.
Why the Timing Matters for the IPO
Anthropic is confidentially preparing for a 2027 IPO alongside OpenAI, and is simultaneously defending a copyright lawsuit from Sony Music Publishing and Warner Chappell that names its founders personally. A public company doesn't get to manage its relationship with federal regulators as a side conversation -- procurement risk, regulatory posture and litigation exposure all become disclosed risk factors underwriters have to price. Reduced friction with the administration overseeing AI policy, federal AI procurement and antitrust enforcement is a genuine business input heading into that process, not just a favorable news cycle.
What's still unresolved: neither Lutnick nor Anthropic specified what "done what we asked" actually means in terms of the original guardrails -- whether Anthropic loosened its restrictions on Pentagon use of Claude, or whether the Defense Department accepted Anthropic's original terms and the friction was more about tone than substance.
What Other Labs Take From This
The reconciliation also sets an implicit precedent for how other AI labs navigate the same tension between safety commitments and federal relationships. OpenAI, Google DeepMind and Meta have all had their own friction points with the administration over model access, procurement terms or safety disclosure -- Demis Hassabis's public call this week for a US pre-deployment testing body, made at the same G20 summit where Lutnick's Anthropic comments landed, is itself a data point in that ongoing negotiation. If Anthropic's path back into the administration's good graces ran through quiet concessions on military-use guardrails, that's a template other labs can either follow or explicitly avoid, depending on how much each one is willing to trade safety posture for smoother federal relationships heading into a period when procurement contracts and regulatory goodwill both carry real commercial value.
