OpenAI, Anthropic, and Google discuss safety as the AI race continues
frontier AI
The most advanced and powerful AI systems in development.
third-party evaluator
An outside group that checks an AI company's safety work.
antitrust law
Law that can stop companies from weakening competition through cooperation.
What happened
OpenAI, the company behind ChatGPT, has been talking with Anthropic, the company behind Claude, and Google DeepMind, Google's AI research group. Chris Lehane, OpenAI's global policy chief, told reporters on September 15 that the talks had continued for several weeks. TechCrunch reported the disclosure.
This confirms contact, not a finished three-way safety agreement. The report gives no final plan.
Why now
The discussion follows a weekend essay by Dario Amodei, Anthropic's CEO. He urged the industry to slow the pace of frontier AI to avoid catastrophic risks. Frontier AI means the most advanced systems in the race.
Sam Altman, OpenAI's CEO, and Demis Hassabis, who leads Google DeepMind, supported the call. Altman also said OpenAI would join Anthropic in embedding third-party evaluators. These evaluators would work from outside the companies while checking their safety practices.
Why it matters
The three companies compete intensely. Their safety talks therefore stand out. A shared approach could help them spot risks that one company misses. It could also make safety promises easier to compare. These are possible benefits, not results confirmed by this report.
The idea of outside review matters for another reason. Safety is not only about how a model performs during a public test. It is also about whether a company can show how it checked the system and reported problems. The article does not provide that deeper evidence.
The competition problem
Cooperation creates a legal problem too. If companies coordinate in ways that weaken competition, antitrust law could become an issue. Amodei proposed a narrow government waiver for safety talks. Lehane reportedly said the companies did not need one.
The political climate adds another conflict. President Donald Trump has called AI safety concerns a hoax. He argues that slowing development would give China an advantage. His AI adviser David Sacks has also called existential-risk fears overblown.
At the same meeting, Lehane backed a FRONTIER Act provision, a proposed rule for leading AI labs. It would require those labs to admit independent verification organizations.
What the report confirms
The strongest confirmed point is OpenAI's admission of weeks-long talks. The report also records support for outside evaluators. It cites separate reporting about a possible industry standards body. It does not say that the body exists or that the three companies have adopted common rules.
What remains unclear
We do not know the talks' agenda, participants, schedule, or outcome. We also do not know which evaluators would enter the companies, what they could inspect, or whether they could publish findings. TechCrunch said it sought comments from Anthropic, Google, and OpenAI.
What to watch next
The next test is implementation. Watch for a public safety framework, clear evaluator access, and reports that companies cannot freely edit. Also watch the FRONTIER Act and any commitment from Google DeepMind. The bigger question is simple: can rivals share safety work without turning cooperation into a new barrier against competition?
Three AI rivals talk about staying safe
📰 Full story: OpenAI, Anthropic, and Google discuss safety as the AI race continues
OpenAI, Anthropic, and Google DeepMind discussed AI safety for weeks.
frontier AI
The most advanced AI systems in the race.
outside evaluator
A group outside the company that checks safety.
💡 The gist
- OpenAI, Anthropic, and Google DeepMind discussed AI safety for weeks.
- Outside groups may check the companies' safety work.
- No shared safety agreement has been announced.
OpenAI is the company behind ChatGPT. Anthropic is the company behind Claude. Google DeepMind is Google's AI research group.
On September 15, Chris Lehane, OpenAI's policy chief, described the talks. TechCrunch reported his comments. AI safety means checking whether powerful AI could cause serious harm.
The talks followed a call from Dario Amodei, Anthropic's CEO. He asked companies to slow the race for frontier AI. Frontier AI means the most advanced AI systems.
Sam Altman, OpenAI's CEO, supported the idea. Demis Hassabis, who leads Google DeepMind, also supported it. Altman said OpenAI would welcome outside evaluators with Anthropic.
An outside evaluator checks safety without working inside the company. This could help find problems a company misses itself. It could also make safety claims easier to trust. But this benefit is only a possibility.
The report does not describe a finished plan. It does not explain the meeting's topics. It does not name the outside evaluators. It does not say how much access they would receive.
Companies also worry about antitrust law. That law can punish cooperation that weakens competition. Amodei suggested narrow government permission for safety talks. Lehane said the companies did not need that permission.
President Donald Trump has rejected the need for slower development. He said slowing down could let China gain an advantage. His adviser David Sacks called extreme AI fears overblown.
OpenAI also supports a FRONTIER Act provision. The proposed rule would require leading AI companies to allow independent checks.
The next question is whether talks become public rules. Watch for commitments from Anthropic and Google DeepMind. Also watch whether outside groups can publish their findings. TechCrunch reported that the talks were still developing.
Big AI companies talked about staying safe
📰 Full story: OpenAI, Anthropic, and Google discuss safety as the AI race continues
OpenAI, Anthropic, and Google DeepMind talked about AI safety.
outside checker
A person or group that looks for dangerous problems from outside.
OpenAI is the company that makes ChatGPT. Anthropic is the company that makes Claude. Google DeepMind is Google's AI team. AI means computer tools that can do tasks. These companies make powerful AI tools. They compete to make better tools. They also talked about staying safe.
Safety means stopping the tools from causing big trouble. Dario Amodei leads Anthropic. He asked companies to slow down. Sam Altman leads OpenAI. He liked the idea. Outside checkers may look at their safety work. They would look for dangerous problems.
The talks are not a new rule yet. The companies have not shared many details. People will watch what happens next. Source: TechCrunch