OpenAI asked Congress if an industry-wide AI slowdown is even legal
The question nobody in Washington wanted to field came from the companies themselves: is it against the law for AI labs to agree to slow down together? According to Wired's Maxwell Zeff, OpenAI has privately asked members of Congress for guidance on whether orchestrating an industry-wide slowdown in AI development would violate antitrust law — the first time a frontier lab has effectively sought a legal safe harbor for the thing safety advocates keep demanding. OpenAI did not respond to Wired's request for comment, so treat the report as sourced, not confirmed.
The problem is real. Coordination on safety — shared incident disclosure, joint pause protocols, common eval standards — is exactly the kind of competitor collaboration antitrust law was built to police. Separately, each lab slowing itself down just hands speed to its rivals, so nobody moves alone. OpenAI asking for explicit legal cover is the system working: the company that would bear the risk is refusing to act until the rules say it can.
Congress has already sketched an answer. In July a bipartisan group introduced the Collaboration on Adversarial Threats and Security Risks Act, which would explicitly permit AI labs to coordinate on security and safety work without antitrust exposure. It has been stuck in the House Judiciary Committee since. Until something like it passes, "we should all slow down" stays legally radioactive — which means every call for collective restraint is, today, a call for Congress to move first.
Fanuc and Google put an AI welding agent on the factory floor
Japan's Fanuc announced an "AI Welding Agent" developed with Google that welds directly from an engineering drawing — zero setup, zero robot teaching. The company positions it as physical AI for arc welding, running on the robot controller itself with no extra hardware, and says user data stays protected on the machine.
The target is the global shortage of skilled welders, and the design bet is that the hard part isn't the weld — it's the programming. If a system can read a blueprint and produce a correct weld without weeks of teach pendants and fixture work, high-mix, low-volume shops become automatable for the first time. That is a bigger deal than another point of speed on a production line: welding is where robot adoption has stalled for decades precisely because the setup cost dwarfs the weld time.
Worth noting: this is Fanuc's second Google collaboration this year, after a May partnership on AI-agent-powered robots, so this looks like a product line, not a one-off demo.
What to watch: whether the House takes up the antitrust safe harbor bill — and whether Fanuc publishes real deployment numbers for the welding agent at trade shows this fall.
When AI labs ask permission to slow down and robots ask to take over the weld, which do you think ships first? Tell us in the comments.
Sources: Wired · Techmeme archive · Fanuc news release · Nikkei Asia
