Google is in active talks to deploy Gemini inside classified Pentagon systems after Anthropic's Claude phase-out. An analytical breakdown of the governance dispute, legal battles, and strategic capital implications reshaping defence AI procurement in 2026.

Key Highlights

  • The Pentagon designated Anthropic a "supply chain risk" after contract talks over Claude's military use collapsed in early March 2026.
  • Google is in active negotiations to deploy Gemini inside classified Department of Defense systems, as first reported by The Information.
  • Anthropic CEO Dario Amodei met White House Chief of Staff Susie Wiles on April 17, signalling a potential diplomatic thaw over the new Mythos AI model.
  • A federal judge temporarily blocked the government's blanket blacklist of Anthropic, though the DC Circuit Court of Appeals declined to bar the Pentagon from severing ties.
  • The dispute is reshaping how AI vendors structure classified deployment contracts, with governance language now a central commercial variable.

A Governance Conflict, Not a Contract Dispute

What has unfolded between Anthropic and the United States Department of Defense since early 2026 is not, at its core, a procurement disagreement. It is a structural contest over who holds the authority to define the permissible limits of military AI.

The fracture became public in late February. Anthropic CEO Dario Amodei reiterated his refusal to allow Claude to be used for mass surveillance of U.S. citizens or to guide fully autonomous weapons, rejecting Pentagon requests to make unfettered use of the model. The Pentagon's position, was that it only sought to use Anthropic's model for "all lawful purposes," dismissing Amodei's stated concerns as a false narrative.

The gap between those two positions proved unbridgeable. The Pentagon went on to declare Anthropic a "supply chain risk," a label only used in the past for companies associated with foreign adversaries, which would effectively blacklist Anthropic from the government. Anthropic filed suit. A federal judge in California subsequently blocked federal agencies outside the Department of Defense from applying the designation. Yet the DC Circuit Court of Appeals, in a separate proceeding, declined to bar the Pentagon itself from cutting ties with Anthropic, noting that intervening "would force the United States military to prolong its dealings with an unwanted vendor of critical AI services in the middle of a significant ongoing military conflict."

The legal scaffolding remains contested. The strategic implications, however, are already visible.

The Gemini Opportunity

Eight years after walking away from a Pentagon drone-analysis contract under employee protest, Google is back at the table. Negotiators from the company are working with the Department of Defense to deploy Gemini inside classified systems, stepping into the opening left by the Anthropic phase-out.

In March 2026, the Pentagon launched GenAI.mil and made Google's Gemini the first model available on its unclassified networks. The classified deployment discussions, first reported by The Information, represent a material escalation of that relationship. People with knowledge of the discussions told that progress has been made toward a deal. The negotiations have reportedly included Google's proposal of language in the contract preventing its AI models from being deployed for autonomous weapons and domestic mass surveillance without human oversight.

This is where the strategic distinction between Anthropic and Google becomes commercially instructive. Anthropic maintained absolute restrictions. Google is pursuing conditional restrictions, namely prohibitions on specific applications paired with broad access across other lawful uses. Defense officials, however, have pushed for broad "all lawful uses" wording that they argue is necessary to move quickly during crises. This tug-of-war over usage guardrails could reshape how the federal government writes AI contracts going forward.

For investors tracking Alphabet's institutional revenue exposure, the classified deployment tier represents a meaningful expansion in addressable market. Such a shift would mark the deepest penetration yet of a commercial frontier model into U.S. classified infrastructure. Cloud revenue implications compound this further, since Google's Gemini runs on its own infrastructure, meaning a classified deployment agreement carries significant compute billing alongside the licensing component.

Vendor Architecture Is Shifting

The Pentagon's broader procurement posture has evolved beyond single-vendor reliance. In recent months, the Defense Department has granted different forms of access to multiple companies, reflecting a desire to avoid reliance on a single provider. Anthropic's Claude, while technically the most capable model on classified networks prior to its phase-out, was also the most architecturally constrained from the Pentagon's perspective due to its usage policy conditions.

One key reason Claude had been the only frontier model deployed on classified networks was that Anthropic's tools were the easiest to deploy on cloud networks powered by AWS, which contributes the largest chunk of the Pentagon's Joint Warfighting Cloud Capability. That infrastructural convenience helped Claude achieve early classified deployment. It also created a single point of fragility once the governance conflict emerged.

The restructuring now underway is multi-vendor by design. According to an Axios report, the military currently can use Grok, Gemini, and OpenAI's ChatGPT for unclassified tasks, and has been negotiating with Google and OpenAI to bring its technology over for use in classified work. OpenAI's emerging contract terms are reported to mirror Google's conditional approach, accepting restrictions on autonomous weapons and mass surveillance while preserving broad operational applicability.

Anthropic: Blacklisted but Not Irrelevant

Anthropic's position, while structurally weakened in the near term, is not static. Two developments in April 2026 suggest the company retains strategic leverage.

The first is Mythos, a new AI model announced in early April that specialises in identifying software vulnerabilities. The White House's interest in the technology comes from the capability of Mythos to dig out thousands of vulnerabilities in commonly used software to create potent exploit chains. The dual-use nature of this capability, powerful for defensive cybersecurity and equally potent as an offensive tool, has drawn the federal government back into dialogue with Anthropic regardless of the ongoing legal conflict over Claude.

Amodei met with Susie Wiles, the White House chief of staff, and other senior administration officials to talk about the company's powerful new Mythos model. The White House said talks were "productive and constructive" and addressed innovation and safety. Treasury Secretary Scott Bessent was also present, reinforcing the economic dimension of the discussion alongside the security framing.

The meeting took place as the U.S. government is trying to balance its hardline approach to Anthropic with the national security implications of turning its back on the company's breakthrough technology, including Mythos, which Anthropic has been rolling out under Project Glasswing to a select group of companies for cybersecurity assessment purposes.

The second source of leverage is litigation. Anthropic's legal challenges remain active. A complete resolution will require either a negotiated contract framework or a judicial determination on the scope of the supply chain risk designation. Neither outcome is imminent.

Capital Market Implications

The defence AI market is not a discrete segment. It spans intelligence analysis, logistics optimisation, decision-support infrastructure, and cybersecurity, each carrying multi-year contract cycles and high switching costs. Firms that establish early classified deployment relationships gain compounding advantages in integration depth, institutional familiarity, and renewal probability.

Anthropic's exclusion from this market, even temporarily, narrows its institutional revenue exposure at a moment when the company is capital-intensive and not yet profitable. The offsetting narrative is that its principled governance stance may strengthen its positioning in regulated enterprise sectors, where reputational differentiation from less constrained AI vendors carries commercial value.

For Google, the potential Gemini classified deployment represents both a revenue opportunity and a reputational recalibration after years of internal debate about defence work. Learning from past controversies such as Project Maven, Google is approaching defence engagement with a stronger emphasis on governance. Whether that governance framing survives the Pentagon's preference for broad operational flexibility remains the central uncertainty in the negotiation.

The episode also carries a structural signal for the broader AI industry. Governments are converging on a preference for what might be described as negotiated AI, conditional access with defined guardrails, rather than either unconditional access or principled refusal. The commercial viability of each governance model, across different institutional client categories, will be a defining strategic variable for AI vendors over the next five years.