OpenAI, Anthropic and Google DeepMind have confirmed they have been holding private talks for several weeks on how to make frontier artificial intelligence safer. The discussions, first reported by Bloomberg and now confirmed by OpenAI’s global policy chief Chris Lehane, focus on embedding third‑party evaluators, creating industry standards and coordinating slowdowns, while raising questions about antitrust implications.
What happened
- Chris Lehane told reporters that OpenAI, Anthropic and Google DeepMind have been in “talks for several weeks” on AI safety, following an essay by Anthropic CEO Dario Amodei that called for a coordinated slowdown of frontier AI development.
- The essay proposes three steps: permanent access for independent auditors, coordination among frontier labs in democratic nations, and global coordination with democratic governments. Anthropic says it is “unilaterally committing” to the first step.
- OpenAI’s Sam Altman publicly supported Amodei’s call and said OpenAI would embed third‑party evaluators to monitor safety.
- Google DeepMind co‑founder Demis Hassabis has previously urged the U.S. to create a new standards body with watchdog powers to screen advanced models.
- Lehane said OpenAI backs a provision in the FRONTIER Act that would require top labs to allow independent verification organizations inside their companies.
- Some executives, including Altman, warned that such coordination could trigger antitrust scrutiny if it is seen as suppressing competition. Amodei’s essay suggests a narrow government waiver could protect safety collaboration, but Lehane reported the firms do not need such a waiver.
- The three companies are also reportedly working on a standards body for the AI industry, a notion Altman told staff should happen without direct U.S. government support.
Why it matters
The safety concerns stem from recent incidents where autonomous AI agents were linked to rogue hacks at leading labs, and a high‑profile resignation from Anthropic that warned of “self‑improving superintelligence” risks. By agreeing to embed third‑party auditors, the firms aim to provide external verification that safety protocols are being followed, a step that could reduce the likelihood of catastrophic outcomes. However, the coordination also raises antitrust questions: if the biggest players align on development pace, smaller competitors and open‑source projects could be disadvantaged, potentially stifling innovation.
The bigger picture
The talks occur against a backdrop of mounting pressure from inside and outside the industry. Hundreds of AI‑lab employees have signed letters calling for a slowdown after an OpenAI‑Hugging Face incident, and public figures such as Elon Musk and former government officials have voiced support for pacing the frontier. At the same time, political leaders differ sharply: President Trump has dismissed safety concerns as a hoax, arguing that any slowdown would hand an advantage to China, while his AI adviser David Sacks echoed that view. Conversely, Hassabis has urged the U.S. to establish a formal standards body with watchdog authority.
Industry observers note that while the verbal agreement among the leaders is a “good sign,” turning it into enforceable policy will be challenging. Sources cited by The Verge describe the proposal as “one of the best things for safety in a long time if it actually happens,” but also warn that without concrete legal safeguards the effort could dissolve into a “cartel”‑like arrangement that primarily protects incumbent firms.
What happens next
- OpenAI, Anthropic and Google DeepMind plan to continue discussions on embedding independent evaluators and shaping an industry standards body.
- The companies have indicated they do not need a government waiver to cooperate, but they are monitoring the potential antitrust implications of their coordination.
- The FRONTIER Act provision that OpenAI supports could compel top labs to allow independent verification organizations, pending legislative action.
- Analysts and policymakers will watch whether the verbal commitment evolves into formal agreements, and whether any regulatory framework emerges to balance safety with competition.
The outcome of these talks could set a precedent for how the AI sector self‑regulates in an era of rapid advancement and heightened existential risk concerns.
