OpenAI, Anthropic, Google Confirm Weeks Of AI Safety Talks logo

OpenAI, Anthropic, Google Confirm Weeks Of AI Safety Talks

OpenAI's global policy chief confirmed the company has been coordinating with rivals Anthropic and Google DeepMind on AI safety measures for weeks, following Dario Amodei's public call for a coordinated development slowdown.

TC
By the AI Desk
Edited by Trace Cohen · Early-stage VC & angel · Founder, New York Venture Partners
2 min read
ShareXLinkedInEmail

THE RUNDOWN

1

Direct safety coordination between OpenAI, Anthropic and Google DeepMind -- normally fierce competitors for talent, compute and customers -- is a meaningfully different posture than the public sniping the three labs have engaged in over the past year.

2

The talks follow Amodei's 'Pace the Frontier' essay calling for recursive self-improvement to be approached very carefully, which OpenAI's Sam Altman and Google DeepMind's Demis Hassabis both publicly cosigned.

3

FTC Chair Andrew Ferguson has already warned publicly that labs seeking both new AI regulation and antitrust protection to coordinate should raise 'alarm bells' -- meaning this coordination is happening under active regulatory scrutiny, not in a vacuum.

4

OpenAI explicitly stated it does not believe a formal antitrust waiver is necessary for the three labs to discuss safety, a notably different legal read than the exemption Amodei's essay seemed to invite.

TC

The VC Read · Trace's Take

Trace Cohen

Three labs that spend most of the year poaching each other's researchers and undercutting each other's API pricing are now coordinating on safety without a formal antitrust waiver -- that gap between 'coordinating' and 'seeking legal cover to coordinate' is exactly what FTC Chair Ferguson flagged, and it's the detail worth tracking closest. If OpenAI is right that no waiver is needed, expect a formal complaint or inquiry to test that read within the next few months; if OpenAI is wrong, this becomes a much bigger regulatory story than three companies talking.

Analysis

OpenAI's global policy chief said the company has been in safety-focused discussions with rivals Anthropic and Google DeepMind for several weeks, according to Bloomberg and corroborated by TechCrunch. The disclosure follows Anthropic CEO Dario Amodei's roughly 3,800-word essay, "We Must Pace the Frontier," published on his personal site September 12, which Pulse covered at the time, arguing that recursive self-improvement -- AI systems designing or training their own successors -- must be "pursued very carefully, if at all."

What's Actually Being Coordinated

The companies have not detailed a specific joint safety framework or shared testing protocol publicly. What's confirmed is that senior policy and safety staff across the three labs have been talking, and that Amodei's public call for slower frontier development was cosigned by OpenAI's Sam Altman and Google DeepMind's Demis Hassabis, along with xAI's Elon Musk -- an unusually broad show of public agreement among labs that otherwise compete aggressively on model releases, pricing and talent.

## What's Actually Being Coordinated The companies have not detailed a specific joint safety framework or shared testing protocol publicly.

The Regulatory Tension

FTC Chair Andrew Ferguson addressed this exact dynamic in Georgetown remarks the same week, warning that AI companies simultaneously lobbying for new regulation and an antitrust exemption to coordinate should make "all of my alarm bells go off." OpenAI's response is notable: the company said it sees no need for a formal antitrust waiver for the three labs to discuss safety, a position that, if untested by regulators, could either become the accepted norm for pre-competitive safety coordination or invite exactly the scrutiny Ferguson previewed.

Competitive Context

This coordination sits awkwardly alongside continued public tension between the same labs. Pulse has covered the OpenAI-Anthropic safety standoff over the Astra model in August, where the two labs publicly disagreed on release-pacing philosophy even as they now describe themselves as coordinating. Meta is notably absent from the Amodei-Altman-Hassabis-Musk cosigning, having taken a more permissive public posture on AI development speed -- a split Pulse has tracked as Zuckerberg lobbies against a unified AI standards body that the other labs have shown more openness toward.

What It Means

For founders and investors in AI-safety-adjacent startups -- eval companies, red-teaming shops, alignment researchers -- explicit coordination among the three best-funded labs is a tailwind: shared safety standards typically create demand for third-party verification and tooling, not less. For the labs themselves, the risk is that "coordination" without a legal safe harbor could later be characterized as anticompetitive behavior if a regulator decides three dominant AI companies agreeing to slow a specific capability counts as market-allocation behavior rather than good-faith safety practice. No formal complaint has been filed, and Ferguson's remarks were a general warning, not directed at this specific disclosure -- but the timing puts real regulatory stakes under what the labs are describing as a voluntary safety conversation.

ShareXLinkedInEmail

Key Sources

2 sources

Reported by Bloomberg · Analysis by Value Add Pulse.

← Back to Pulse

THE WIRE in your inbox— Tech, startup & VC news with Trace's take. Free, no spam.