Leading frontier AI laboratories Anthropic, OpenAI, and Google have discussed forming a collaborative industry standards body, according to The Information, as researchers inside the same companies issue stark public warnings about the unchecked race toward superintelligence.
The talks come amid alarming statements from key safety figures. Anthropic researcher Jacob Coxon said: 'Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives.' His alignment lead Evan Hubinger publicly agreed, adding that AI could kill all humans and that he personally puts the probability above 10% within the next decade. OpenAI’s chief scientist Jakub Pachocki separately stated that no lab has sufficiently solved alignment and monitoring to justify continued maximum-speed scaling.
That existential anxiety has reached the highest political levels. In the UK Parliament on Thursday, Sir Ed Davey MP pressed Prime Minister Andy Burnham over a Financial Times report that Anthropic did not submit its latest model to the UK Institute for Testing, allegedly due to pressure from the Trump administration. The Prime Minister acknowledged AI risks to national security but offered no concrete brake on the race, saying AI could also be 'a source of solutions to keep us safer.'
The industry-led standards effort is not new. On July 14, Demis Hassabis, head of Google DeepMind, proposed a U.S.-based Frontier AI Standards Body modeled on FINRA, under which frontier labs would voluntarily submit advanced technologies for safety assessment 30 days before release. Dario Amodei, CEO of Anthropic, argued in his article 'We Must Pace the Frontier' that independent evaluators such as METR should be allowed extended access inside frontier laboratories.
Earlier coordination attempts include the 2023 Frontier Model Forum, backed by six major AI companies and a $10 million AI Safety Fund; the Agentic AI Foundation launched in December 2025 under the Linux Foundation; and the Appia Foundation created in June 2026 to turn AI governance principles into specifications and compliance proofs. The EU’s General-Purpose AI Code of Practice already counts Anthropic, Google, Microsoft, and OpenAI among its signatories.
OpenAI’s head of global affairs Chris Lehane has called for the US and China to support a global AI safety framework, comparing the initiative to international nuclear cooperation.
Some commentary frames the current warnings as a second alignment failure: the first was the rise of limited liability corporations optimized solely for profit under the Friedman doctrine, an experiment that already outpaced regulatory control. From that perspective, the AI race is not unprecedented in kind, but only in speed and scale.
Analysts warn that an issuer-pays model could recreate conflicts similar to the credit-rating failures before the 2008 financial crisis. The Council on Foreign Relations has questioned regulator access to frontier models, evaluator expertise, national security safeguards, and testing reliability. Capacity remains another challenge: GovAI projects 14–16 models between 2025 and 2028 could match the scale of the largest training runs to date, requiring continuous review rather than one-off assessments. There are also competition concerns that compliance costs fixed by the biggest labs could become entry barriers for startups.
