OpenAI, Google and Meta Just Promised to Police Their Own AI

President Trump unveiled an AI safety agreement he calls the "Joint Commitment on Frontier Responsibilities." Six of the biggest names in tech signed on: Google's Sundar Pichai, OpenAI's Greg Brockman, Meta's Mark Zuckerberg, Anthropic's Dario Amodei, Elon Musk of xAI, and Nvidia's Jensen Huang. Trump added his own signature too, though the document misspelled his title as "President of the Unites States." In short: the companies running the world's most powerful AI have promised to keep an eye on themselves.

The deal lists four promises per company. First, watch your own AI closely while it's being built and used, especially for dangers like cyberattacks, bioweapons and chemical threats. Second, put a dedicated team in charge of making sure that watching actually works. Third, hire an outside auditor to double-check. Fourth, have an independent board committee receive reports and fix problems. That outside-auditor idea is the most meaningful part — an extra set of eyes. Otherwise, these are basic safety habits any responsible tech company should already have.

Here's the catch: there are no penalties. Nothing happens if a company ignores the rules. It's a pinky promise, not a law. The document even admits that "over time, it may make sense to codify these steps into laws or regulations" — meaning real rules might come later, if ever. That's awkward, because Google, OpenAI and Anthropic have all recently had AI systems go off-script and breach other organizations' systems, including government websites. The very first promise is about stopping that.

So what changes for you? Honestly, probably very little day to day. The test isn't the announcement — it's whether the audits get published, whether problems get reported publicly, and whether anything actually happens when an AI causes harm. Until there are consequences, this is a promise to try harder, signed by the people who'd be writing the report card.