Grok Goes to War
The day's most explosive claim came from a sworn legal filing: a senior U.S. defense official asserted that the Pentagon used Elon Musk's Grok chatbot to help launch 2,000 missiles at Iran. The disclosure, reported by The Independent, AOL, MSN and Arab News PK among others, marks the U.S. government acknowledging that Grok was used to support military strikes. The detail that an AI chatbot featured in targeting decisions of this scale will dominate the debate over how far automation has already crept into the use of force.
That debate now has a counterweight. A Senate committee overseeing the Pentagon adopted a measure — an amendment from Arizona Sen. Mark Kelly, per KJZZ — aimed at keeping people, not algorithms, accountable when force is used. The timing is hard to ignore: lawmakers are moving to keep humans in charge just as a marquee example of AI in the kill chain surfaces.
The Pentagon's AI-First Turn
The broader backdrop is a military leaning hard into AI. Fast Company reports that Defense Department official Cameron Stanley is pushing the Pentagon to become "AI-first," with the military striking deals with seven tech firms. Separate reporting from defense outlets including DVIDS describes AI being folded into the unglamorous-but-critical work of making decisions and shipping software faster. The message: AI is being treated less as a futuristic add-on and more as core infrastructure for how the department operates.
The hardware side is moving too. The Air Force tapped Anduril and General Atomics to build its first operational Collaborative Combat Aircraft drones, DefenseScoop reports, pushing the program past design and toward fielded aircraft. Feeding the software behind such systems: more than 500,000 hours of drone footage from Russia's war against Ukraine have been added to a military AI training dataset, according to the Kyiv Post and UNITED24 Media — a sign of how real-world combat data is becoming fuel for the next generation of military models.
Washington Tightens the Screws on Anthropic
In a striking twist, Anthropic — the maker of Claude, and a company that has spent years publicly asking to be regulated — is now on the receiving end of government restrictions. CNBC reports the Trump administration is responding in a way the company didn't welcome. Per Memeburn, the White House moved to curb access to Anthropic's most capable models after Amazon raised concerns. A firm that lobbied for guardrails is finding the guardrails pointed at its own most powerful systems.
Governance Tries to Catch Up
The governance gap loomed over two more stories. The Bulletin of the Atomic Scientists argues AI is reshaping the risk landscape around biological agents and nuclear weapons — and asks whether oversight can keep pace. And in the UK, Wired reports the government plans to use facial age-estimation technology to help judge how old asylum seekers are, pressing ahead even though its own internal testing flagged serious problems with the approach. It's a cautionary case of deploying AI despite known flaws.
A New Hub Wants In
Finally, a note on geography. At a robotics and AI forum at Carnegie Mellon University on Wednesday, regional stakeholders pitched Pittsburgh as a defense tech hub — an effort to recast the old steel city as a center of gravity for the very technologies dominating today's headlines.
The through-line: AI is no longer adjacent to defense decisions, software, and weapons — it's inside them. The open question, raised everywhere from the Senate floor to a courtroom filing, is who stays accountable when it is.