The marriage between Silicon Valley and the Pentagon has never been simple. It’s gotten a lot more complicated.
Over the past several months, a growing number of artificial intelligence companies have moved aggressively to secure defense contracts, seeking to embed their technology in military operations ranging from logistics optimization to intelligence analysis. Anthropic, the San Francisco–based AI safety company backed by billions in venture capital and a major investment from Amazon, has been among the most closely watched players in this space — and among the most politically exposed. Congressional scrutiny of AI companies pursuing defense work has intensified, with lawmakers on both sides raising questions about conflicts of interest, foreign investment entanglements, and the risks that come when companies simultaneously court the national security establishment and global commercial markets.
Senator Elizabeth Warren has been one of the loudest voices pressing these concerns. The Massachusetts Democrat has a long record of questioning the revolving door between government and industry, and she’s turned that lens sharply on the AI sector. Warren has raised pointed questions about whether AI firms doing business with the Department of Defense face inherent supply chain risks — particularly when their investors, cloud computing partners, or overseas business interests could create pressure points that adversaries might exploit. Her inquiries have also touched on a more uncomfortable question: whether companies that speak publicly about AI safety are using that rhetoric as a shield while pursuing lucrative military work with insufficient oversight.
These aren’t abstract concerns. The Pentagon’s adoption of AI tools has accelerated under both the Biden and Trump administrations, driven by a bipartisan consensus that the United States must maintain technological superiority over China. The Department of Defense’s Chief Digital and Artificial Intelligence Office (CDAO) has been tasked with integrating AI across military functions at scale. That urgency has created enormous commercial opportunity — and enormous political risk for the companies involved.
Anthropic’s position is particularly nuanced. The company was founded in 2021 by former OpenAI executives Dario and Daniela Amodei, who left in part over disagreements about safety and commercialization. Anthropic has built its brand around responsible AI development, publishing extensive research on AI alignment and establishing itself as a credible voice in policy discussions about the technology’s risks. But the company has also grown rapidly, raising over $7 billion in funding and competing fiercely for enterprise and government customers with its Claude family of AI models.
That dual identity — safety-focused researcher and aggressive commercial competitor — creates tension.
Warren’s scrutiny reflects a broader pattern of congressional concern about the defense-tech pipeline. In recent months, multiple lawmakers have questioned whether adequate guardrails exist for AI procurement at the Pentagon. The worry isn’t just about the technology itself but about the corporate structures behind it. When an AI company’s largest investor is also a major cloud services provider to the intelligence community, as Amazon Web Services is, the interdependencies multiply. When that same company has business relationships spanning multiple countries, including nations with complex geopolitical relationships with the United States, the risk calculus changes further.
Supply chain risk in defense has traditionally referred to physical components — semiconductors, rare earth minerals, weapons systems parts. The AI era introduces a different kind of vulnerability. Software supply chains are harder to audit. Model training data can be opaque. And the companies building these systems often depend on compute infrastructure they don’t own or fully control. A Reuters analysis of AI defense contracting noted that the sector’s rapid growth has outpaced the government’s ability to conduct thorough security reviews of new vendors.
There’s also the retaliation question — one that Warren has raised explicitly. If an AI company becomes deeply embedded in Pentagon operations, does that create leverage that could be used against it? Could a foreign government threaten to restrict market access or impose regulatory burdens on an AI company’s commercial operations as a way to influence its defense work? The scenario isn’t hypothetical. China has used market access as a coercive tool against companies in other sectors, from semiconductors to social media. The question is whether AI companies have adequately gamed out these scenarios and whether the Pentagon has demanded that they do so.
Anthropic has taken some steps to address these concerns. The company has published its responsible scaling policy, which outlines how it evaluates the risks of increasingly capable AI systems before deployment. It has also engaged with policymakers in Washington more actively than many of its peers, participating in voluntary safety commitments and testifying before Congress. But critics, Warren among them, argue that voluntary commitments are insufficient when national security is at stake.
The political dynamics here are layered. Some Republicans have pushed back on what they see as excessive regulatory caution, arguing that bureaucratic delays in AI adoption will hand advantages to China. Former Google CEO Eric Schmidt, who has become one of the most prominent advocates for military AI adoption, has warned repeatedly that the U.S. risks falling behind if it doesn’t move faster. Schmidt’s Defense Innovation Board work and his subsequent ventures in defense technology have made him a polarizing figure — admired by hawks, distrusted by those who see a too-cozy relationship between tech billionaires and the military establishment.
And then there’s the workforce dimension. Many AI researchers entered the field motivated by a desire to build beneficial technology, not weapons systems. Google’s Project Maven controversy in 2018 — when thousands of employees protested the company’s work on Pentagon drone imagery analysis — demonstrated the internal fractures that defense work can create within tech companies. Anthropic has cultivated a workforce deeply invested in the idea that AI safety is the company’s core mission. Pursuing defense contracts, even defensive or logistical ones, tests the boundaries of that culture.
So where does this leave the industry?
The most likely outcome is continued expansion of AI companies into defense, accompanied by escalating congressional oversight. The days when Silicon Valley could treat government work as a side project are over. Defense AI is becoming a core business line for companies that once would have shunned it. That shift brings revenue, but it also brings the kind of scrutiny that tech companies have historically been poorly equipped to handle — scrutiny not just of their products but of their corporate governance, their investor relationships, and their geopolitical exposure.
Warren’s inquiries, whatever their immediate political motivations, point to a structural problem that won’t resolve itself. The U.S. government is becoming deeply dependent on a small number of private AI companies for capabilities it considers essential to national defense. Those companies are simultaneously pursuing global commercial strategies that may conflict with their defense obligations. And the oversight mechanisms designed for traditional defense contractors — companies like Lockheed Martin or Raytheon that derive the vast majority of their revenue from government work — don’t map cleanly onto firms that are trying to be everything to everyone.
The Pentagon knows this. In a Department of Defense briefing earlier this year, officials acknowledged that the acquisition process for AI is fundamentally different from procurement of traditional weapons systems. Software evolves continuously. Models are updated, retrained, fine-tuned. The notion of a fixed contract deliverable doesn’t apply in the same way. That creates accountability gaps — gaps that lawmakers like Warren are eager to fill, and that defense hawks worry will be filled with red tape rather than real security improvements.
Anthropic isn’t the only company facing these pressures. OpenAI has also moved into defense-adjacent work, as has Palantir, which has long straddled the line between Silicon Valley and the national security world. But Anthropic’s particular combination of safety branding, massive private investment, and growing government ambition makes it a focal point for the debate. The company’s next moves — and Congress’s response — will set precedents that shape how AI integrates into American defense for years to come.
For now, the tension is unresolved. AI companies want defense dollars. The Pentagon wants AI capabilities. Lawmakers want oversight. And the adversaries these systems are meant to counter aren’t waiting for Washington to sort out its procurement processes. Something has to give. The question is what — and who pays the price when it does.
WebProNews is an iEntry Publication