OpenAI, Google and Anthropic Are Quietly Teaming Up on AI Safety
The rivals building the world's smartest AI are now comparing safety notes.
The three biggest players in AI — OpenAI, Google DeepMind and Anthropic — have been talking behind closed doors for weeks about how to keep advanced AI safe. OpenAI's policy chief, Chris Lehane, let it slip Tuesday while in Washington to meet with lawmakers. The talks follow an essay by Anthropic's CEO, Dario Amodei, calling on the industry to slow down the race toward ever-more-powerful AI. Rival CEOs including OpenAI's Sam Altman and Google's Demis Hassabis publicly backed him. So what? The people who understand this technology best are worried enough to cooperate — that alone is a signal worth noticing.
The companies aren't just chatting. They're reportedly working on a shared standards body — an independent watchdog that could inspect the most powerful AI models and hit the brakes if something looks dangerous. OpenAI says it also supports letting outside 'verification organizations' look inside top AI labs. So what? Until now, AI companies have mostly policed themselves. Outsiders peering in could mean earlier warnings before something goes badly wrong.
There's a big catch. When competitors coordinate, they can run afoul of antitrust law — rules meant to stop companies from fixing prices or squeezing out rivals. Amodei suggested a narrow government waiver; Lehane says they don't need one. Meanwhile, President Trump has called AI safety fears a hoax and warned that any slowdown hands China an advantage. His AI advisor, David Sacks, who has invested across the industry, agrees. So what? The people building AI want guardrails, but the government isn't offering much help.
For you, the stakes are practical. If these companies succeed, the AI tools you use for work, school or health advice could be tested by outside experts before release — a bit like crash-testing a car. If the effort stalls, you'll likely keep seeing fast, powerful, lightly checked AI arrive with little warning. Either way, this week's news shows the industry's biggest names don't fully trust their own creations.
- OpenAI, Google DeepMind and Anthropic have held private safety talks for weeks — unusual for three direct competitors
- They're discussing an independent watchdog to inspect the most powerful AI models and slow things down if risks grow
- Washington is split: Trump calls safety fears a 'hoax,' while some lawmakers push rules requiring outside inspectors inside AI labs
Why It Matters
The people building AI are worried enough to cooperate — that hints at real risks and possible new rules ahead.