OpenAI is discussing possible cooperation on artificial-intelligence safety with Anthropic and Google DeepMind, Bloomberg reported on 15 September, in an account carried by Reuters. Chris Lehane, OpenAI’s global policy chief, said the talks had run for several weeks and that, in his reading, a special antitrust waiver was not needed to coordinate on safety.

Reuters said it had not independently verified the report and that the three companies did not immediately offer a detailed comment. Later an OpenAI spokesperson told other outlets that discussions with rivals, including Anthropic and Google, about common standards were under way. Confirmation of talks is not a signed agreement.

Risk assessment, testing and support for bipartisan legislation aimed at catastrophic harms could be on the table. Collaboration among competitors may raise a common bar, but it leaves open questions of transparency, independent verification and the line between safety and commercial secrets.

Voluntary pledges do not replace public oversight. Until a joint text exists, this article speaks of discussions, not a completed partnership and not an agreed technical pause among labs.

The three firms remain competitors on frontier models. A Washington policy briefing does not, by itself, change release calendars.

The core information belongs to Bloomberg; Reuters marked it as not independently verified. That distinction stays in the text.

The picture is a 2018 technology-conference frame, not the OpenAI briefing of September 2026 and not the signing of a pact.

Image: Web Summit 2018, Centre Stage / Wikimedia Commons (Web Summit, CC BY 2.0). A 2018 conference, not the September 2026 AI-safety talks. Cropped to 16:9.

Source consulted: Reuters — OpenAI is working with Anthropic, Google on AI safety, Bloomberg News reports; OpenAI — Safety.