OpenAI, Anthropic, Google have been in talks on AI safety for weeks

The fact that OpenAI, Anthropic, and Google DeepMind are talking is not itself surprising—they've done this before, most notably in 2023 with the Frontier Model Forum. What's notable is the timing and the political context. These companies are fierce commercial rivals locked in a high-stakes race for dominance, yet they're finding common ground on safety architecture. That's both reassuring and deeply concerning: reassuring because self-awareness about catastrophic risks exists at the highest levels, concerning because the very fact they feel the need to coordinate outside of government channels suggests they don't trust the political system to handle it responsibly. The reported posture of the Trump team—dismissing safety concerns in favor of "keeping pace with China"—creates a bizarre incentive structure. If the government signals that speed is the only metric that matters, then any company that voluntarily slows down to address safety is unilaterally disarming in the global race. This is the classic collective action problem, and it's precisely why the companies appear to be building their own guardrails in parallel. They know that the political winds can shift, but the technology won't wait. What makes these talks particularly interesting is the subtext. OpenAI and Anthropic have very different architectural philosophies—one pushing toward AGI with all deliberate speed, the other prioritizing alignment from day one. Google DeepMind sits somewhere in between, with the deepest research bench. If they can agree on shared safety protocols, it would represent a de facto industry standard that regulators could eventually adopt. But there's also a risk: if these talks produce only vague commitments and press releases, they'll function as a fig leaf that gives politicians cover to do nothing while the real decisions happen in private boardrooms. The deeper question is whether voluntary coordination can outpace the competitive pressures that make safety a secondary consideration. History suggests it's a fragile arrangement. The original Asilomar principles and the various moratoriums on dangerous research were all voluntary too, and they eroded as soon as the commercial stakes got high enough. The fact that these talks are happening at all is a positive signal, but it's a small one. The real test will come the first time a safety protocol collides with a product deadline and a billion-dollar revenue opportunity. Source: [TechCrunch](https://techcrunch.com/2026/09/15/openai-anthropic-google-have-been-in-talks-on-ai-safety-for-weeks/)