Technology & Science
Pentagon Brings Custom ChatGPT Online, Signals Next Step Toward Classified-Level AI Access
Between 11–12 Feb 2026 the Defense Department activated a siloed version of ChatGPT for its 3 million-user GenAI.mil network and, a day later, its CTO told tech firms the Pentagon now wants similar frontier models cleared for classified domains.
Focusing Facts
- GenAI.mil’s ChatGPT instance went live 11 Feb 2026 and is approved only for unclassified traffic, serving roughly 3 million civilian and uniformed personnel.
- On 12 Feb 2026 Pentagon CTO Emil Michael said at a White House meeting that the military is “moving to deploy frontier AI capabilities across all classification levels,” pressing companies to drop many usage restrictions.
- In July 2025 the Department awarded contracts worth up to $200 million to OpenAI, Google, Anthropic and xAI to build defense-focused AI solutions.
Context
From the war rooms of WW II’s 1943 Office of Scientific Research and Development to ARPANET’s 1969 debut, U.S. security agencies have repeatedly absorbed civilian tech once it reached strategic maturity. The ChatGPT adoption echoes those pivots: like early mainframes, today’s large language models migrate from academia to commercial clouds and now to government enclaves, tightening the civil-mil nexus that Eisenhower cautioned about in 1961. The Pentagon’s push to strip guardrails on classified networks underscores a long trend toward automation in command-and-control—from SAGE radar (1958) to Project Maven (2017)—but it also revives old worries of over-delegation and error propagation first seen with the 1980 NORAD false alarms. On a 100-year arc, whether AI becomes as indispensable as GPS (first DoD-only, later universal) or is reined in like chemical weapons after 1925 will hinge on who sets the safety norms now; this week’s twin moves mark a decisive step toward embedding generative AI deep inside the U.S. war bureaucracy, setting precedents other powers will quickly study and emulate.
Perspectives
Defense-friendly mainstream and tech trade press
e.g., The Hill, Cryptopolitan, ForkLog — Frame OpenAI’s ChatGPT rollout on GenAI.mil as a welcome leap that will give U.S. forces cutting-edge tools to deter adversaries and keep a technological edge. Leans into Pentagon press-release language about "protecting people" and "maintaining advantage," largely glossing over civil-liberties or accuracy concerns that could complicate the success story.
Tech accountability–oriented outlet (Decrypt) citing public-interest advocates
Tech accountability–oriented outlet (Decrypt) citing public-interest advocates — Highlights warnings from Public Citizen that soldiers may over-trust large language models, making any sensitive data fed into ChatGPT vulnerable to adversaries. Focuses almost exclusively on risks and user error, potentially underplaying safeguards or the Pentagon’s claim that only unclassified data will flow through the system.
Reuters-syndicated global business media
e.g., Devdiscourse, Yahoo! Finance, U.S. News, CNA — Reports that the Pentagon is pressuring AI firms to place models on classified networks with fewer restrictions, stressing the clash between military demands and company guardrails and the danger of AI hallucinations in lethal contexts. Wire-service pieces aim for balance yet rely on anonymous officials and dramatic battlefield imagery, which can amplify a sense of urgency and controversy to drive clicks.