OpenAI, Anthropic and Google DeepMind held weeks-long safety talks, planned third-party evaluators

OpenAI, Anthropic and Google DeepMind held weeks-long meetings to coordinate AI safety standards and third-party evaluation plans, TechCrunch reports.

Chris Lehane confirmed the talks and said the companies coordinated on standards but do not require a government waiver; the reporting frames the meetings as industry actors aligning practices rather than seeking formal regulatory exemptions.

Anthropic CEO Dario Amodei’s essay urging the industry to slow frontier AI preceded the discussions and functions in the reporting as the immediate impetus for companies to talk about shared guardrails. TechCrunch also notes that observers have raised antitrust concerns about such coordination even as company spokespeople pushed back on the need for regulatory waivers.

The reporting is specific about the topics discussed—standards and third-party evaluation plans—but does not provide an agenda, minutes, or the identities of all participants beyond the three firms named. That leaves open questions about whether the talks will produce binding commitments, voluntary norms, or concrete third-party testing regimes.

If the companies adopt shared evaluation frameworks, proponents argue that third-party evaluation could improve transparency and safety; critics counter that privately coordinated standards could harden market positions and attract antitrust scrutiny.

For now, TechCrunch documents the occurrence and motivations—Amodei’s essay and Lehane’s confirmation—without offering documentary evidence of enforceable agreements or timelines for implementation.