Skip to main content

AI Giants Secretly Discuss Safety Standards Group

AI Giants Secretly Discuss Safety Standards Group

According to insiders, Anthropic, OpenAI, and Google have been quietly meeting since July to establish an AI industry standards organization. The talks gained momentum after Anthropic CEO Dario Amodei publicly called for collaboration on technical testing and audits last Saturday.

Representatives from the three companies—below CEO level—have formed a working group that meets regularly, with the latest meeting just last week. OpenAI CEO Sam Altman voiced support for an industry-led testing and audit body during a company-wide meeting this week. However, he stressed that major AI labs must set up such a body independently, without U.S. government support. Altman had initially hoped for government backing, but consensus remains elusive, with some companies opposing the idea.

Three Models, No Agreement

Amodei's blog post, "We Must Slow Down the Pace of Frontier Development," argued for an independent oversight mechanism, with industry self-regulation as a key component. He proposed that companies implement voluntary safety standards while governments draft AI regulations. He also called on leading global firms to grant external evaluators like METR access equivalent to internal employees, pledging Anthropic would lead by example. Additionally, he urged limited antitrust exemptions to enable safe collaboration.

Demis Hassabis, who recently stepped down as CEO of Google DeepMind, said the proposal was "in the right direction" but some details "still need refinement." He had previously suggested building an AI self-regulatory organization modeled on the U.S. Financial Industry Regulatory Authority (FINRA) in July. Elon Musk, CEO of SpaceX, also expressed support.

According to a previous report by The Information, the White House circulated a draft executive order to establish this regulatory body but failed to gain enough support by late summer, putting the work on hold. David Sacks, former White House AI director, has openly opposed a FINRA-like institution.

'A Cartel's Design'?

Criticism is mounting. Aiden Gomez, CEO of Cohere, warned that large AI companies jointly setting rules could restrict smaller competitors: "This idea sounds exactly like a cartel's design." Sacks questioned whether slowing development was entirely altruistic, arguing that product liability risks already provide economic incentives to prevent harm.

Meta's position remains unclear. Its Chief AI Officer, Alexander Wang, only stated on X that "alignment is the foundation of building a personal superintelligence for all people," without clearly supporting or opposing Amodei's full proposal.

Key Points

  • Secret talks: Anthropic, OpenAI, and Google have been meeting since July to form an AI industry standards group.
  • Altman's stance: Supports an industry-led body but insists it must be independent of government support.
  • Amodei's push: Advocates slowing frontier development, voluntary safety standards, and external evaluator access.
  • Critics speak out: Cohere's CEO calls it a "cartel's design"; others question motives.
  • White House stall: A draft executive order failed to gain traction, leaving industry self-regulation as the main path forward.