Independent reporting on American politics
STATE BEACON

OpenAI, Anthropic and Google DeepMind hold weeks‑long AI safety talks, warn of antitrust risk

OpenAI’s policy chief Chris Lehane confirmed that the lab has been meeting with Anthropic and Google DeepMind for weeks to coordinate safety measures, including plans to embed third‑party evaluators, while executives cautioned the effort could trigger antitrust scrutiny.

By State Beacon·
OpenAI headquarters building in San Francisco

OpenAI, Anthropic and Google DeepMind have been holding weeks‑long discussions to coordinate AI safety, and executives have warned that the collaboration could raise antitrust concerns, according to a TechCrunch report published on 15 September 2026.

Weeks‑long safety talks confirmed

Chris Lehane, OpenAI’s global policy chief, told reporters on Tuesday that the company has been working with rivals Anthropic and Google DeepMind on AI safety for weeks. He said, “Lehane told reporters that the company has been working with rivals Anthropic and Google DeepMind on AI safety for weeks,” a statement that TechCrunch attributes to Lehane’s remarks in Washington (TechCrunch, 15 Sep 2026).

The timing of the talks places their start several weeks before the public confirmation on 15 September 2026, as indicated in the packet’s timeline. No exact start date is given, but the description of “weeks‑long” suggests a sustained engagement rather than a single meeting.

Embedding third‑party evaluators

OpenAI’s chief executive Sam Altman said the lab would join Anthropic in embedding third‑party safety evaluators into its development process. The packet records Altman’s comment as, “Altman said that OpenAI would join Anthropic in embedding third‑party evaluators to monitor for safety.” This plan aligns with OpenAI’s public support for the FRONTIER Act provision that would require top frontier labs to allow independent verification organizations access to their models.

Lehane reinforced that stance, stating, “Lehane said that OpenAI supports a provision in the FRONTIER Act that would force top frontier labs to allow ‘independent verification organizations’ into their companies to ensure models are developed safely.” (TechCrunch, 15 Sep 2026).

Antitrust warning from executives

Both Lehane and other senior executives cautioned that the coordination could expose the firms to antitrust liability. The packet notes, “Lehane and other executives warned that such coordination could expose the firms to antitrust liability.” The warning reflects a broader industry concern that joint safety initiatives might be interpreted as collusive behavior under U.S. competition law.

The antitrust risk is not merely theoretical. By discussing joint safety standards and shared evaluator access, the three labs could be seen as limiting competition in the emerging frontier‑AI market, a sector that regulators have begun to scrutinize more closely.

Company snapshots

Key facts for the three firms involved in the safety talks
Company Chief executive Headquarters Country Employees (reported)
OpenAI Sam Altman San Francisco United States 4,500
Anthropic 2,500
Google DeepMind London United Kingdom 10,000

Employee counts and headquarters are drawn from the packet’s company‑research section, which cites Wikidata as the source. The packet flags these figures as background only and advises confirmation against each firm’s filings before publication.

Political backdrop

The disclosure arrives as President Donald Trump publicly dismisses AI safety risks and urges faster development. That rhetoric heightens political pressure on the industry to demonstrate coordinated safety efforts, making the antitrust warning especially salient. While the packet does not contain a direct comment from the White House, the “why_now” note in the commission highlights the contrast between the administration’s stance and the firms’ precautionary measures.

What remains unknown

  • The precise agenda of the safety talks – the packet confirms only that they cover third‑party evaluator embedding and antitrust awareness.
  • Whether any formal agreement or joint governance structure has been drafted.
  • The identity of the third‑party evaluators that OpenAI intends to embed.
  • How regulators, particularly the Federal Trade Commission, will respond if the coordination is deemed anticompetitive.

OpenAI, Anthropic and Google DeepMind have not disclosed timelines for implementing the evaluator program, nor have they quantified any expected impact on model development speed or safety outcomes.

Implications for the sector

If the three labs move forward with shared safety protocols, smaller AI startups may face a de‑facto standard that they must match to compete, potentially raising entry barriers. Conversely, the coordination could set a benchmark that eases regulatory scrutiny for the broader industry, provided antitrust concerns are addressed.

Stakeholders – from venture investors to policy makers – will be watching how the firms balance safety collaboration with competition law. The next public statements from the companies, or any filing with the FTC, will likely clarify whether the talks evolve into a formal safety watchdog or remain informal exchanges.

For now, the only concrete evidence is the TechCrunch interview with Chris Lehane and the quoted remarks from Sam Altman. As the political climate pushes for rapid AI advancement, the industry’s own safety dialogue, now explicitly linked to antitrust risk, adds a new layer of complexity to the ongoing debate over how to govern frontier AI.