The Pentagon has signed contracts with seven artificial intelligence firms to bring their models inside its classified computer networks, the Washington Post reported on May 1. The deals extend access to advanced large language models well beyond the unclassified systems where most of the federal government's AI experiments have lived so far.
The companies, which include Microsoft, Amazon, Nvidia and Reflection AI, will now have their tools running on the Defense Department's secret-level networks, where intelligence about adversaries, war planning and weapons development is processed and stored. Officials told the Post that the agreements are intended to give analysts and decision-makers faster ways to summarise documents, draft assessments and surface patterns in large volumes of classified data.
A door the AI labs have wanted opened
For the major AI labs, the deals represent something the industry has chased for several years: a route into the parts of government where the most sensitive work happens. Until recently, federal AI rollouts had been confined to administrative tasks, customer-facing chatbots, and unclassified cybersecurity tooling. Bringing models behind the secret-level firewall is a meaningful step up.
The arrangement also raises the kinds of questions that have followed every other deployment of large language models in high-stakes settings. Hallucinations are still a known failure mode, and intelligence work is unforgiving of confident-sounding errors. The Pentagon has said the systems will assist analysts rather than replace them, and outputs will be reviewed by humans before any action is taken. Whether that workflow holds up in practice, especially under operational pressure, will be tested on a different timeline than contracts.
Part of a broader thaw with national security
The shift comes alongside a separate report that the National Security Agency has been testing Anthropic's latest model to find cybersecurity vulnerabilities in popular software, including Microsoft products. Taken together, the moves point to a warming relationship between the leading AI labs and the U.S. national security apparatus, one that was distant and at times tense as recently as 2023.
It also fits with the White House's National Policy Framework for Artificial Intelligence, released in March, which urged faster federal adoption of AI while pushing back on a patchwork of state-level rules. Industry executives have argued for some time that government use is the proving ground that matters most, and the latest Pentagon agreements suggest the door is finally open.
For now, the practical questions are about integration and trust. The Defense Department has invested heavily in the infrastructure to run these models inside its secure environments, and the AI firms have invested in clearances and dedicated staff. The harder work, and the more interesting story to watch, is whether the tools end up changing how intelligence is produced rather than simply speeding up what already happens.
Commentarii · 0