Stablecoins
Google, OpenAI, and Anthropic Aim to Establish AI Safety Standards Body by 2027
Google DeepMind, OpenAI, and Anthropic are in discussions to form an independent, industry-led organization focused on establishing AI safety standards. This initiative, which began in July 2026, aims to create shared protocols for evaluating advanced AI models and conducting independent safety assessments before these systems are released to the public.
Anthropic CEO Dario Amodei has emphasized the need for external evaluators in AI safety, arguing that the rapid pace of AI development outstrips the ability of any single entity to manage safety implications. OpenAI CEO Sam Altman has echoed this sentiment, highlighting the necessity of independent standards in light of slow government regulation.
Despite their intentions, the three companies face challenges, as they received C+ or lower ratings on the 2026 Future of Life Institute AI Safety Index, which assesses transparency and responsible development practices. Their previous collaboration, the Frontier Model Forum, aimed at advancing AI safety research but lacked the structure proposed for the new standards body.
As of mid-September 2026, no formal organization has been established, raising concerns about the execution of this initiative. Smaller AI firms should be aware that if the major players succeed in creating these standards, compliance costs could disproportionately impact them, potentially leading to further consolidation in the industry.
New Developments in AI Reasoning
- Thore Graepel, co-creator of AlphaGo, has left Google DeepMind to establish a startup focused on AI reasoning.
- He is currently raising initial funding in the tens of millions, with future rounds expected to reach hundreds of millions.
- Graepel aims to teach machines to think through problems, moving beyond simply increasing the size of language models.
- His new venture emphasizes structured search methods and probabilistic planning, contrasting with traditional language model approaches.
- Graepel's startup targets the gap between fluency in language and actual competence in problem-solving.
- 2026 has seen a notable exodus of senior talent from DeepMind, including David Silver, another architect of AlphaGo, who founded Ineffable Intelligence.
- The trend of leadership changes at DeepMind and the startup boom in AI funding is contributing to this brain drain.
- Probabilistic planning, a key aspect of Graepel's approach, explicitly deals with uncertainty in decision-making processes.
FAQ
What is the purpose of the proposed AI safety standards body?
The proposed AI safety standards body aims to establish shared protocols for evaluating advanced AI models and conducting independent safety assessments before these systems are released to the public.
Who are the main companies involved in this initiative?
The main companies involved in this initiative are Google DeepMind, OpenAI, and Anthropic.
Why do the companies believe independent evaluators are necessary?
The companies believe independent evaluators are necessary because the rapid pace of AI development outstrips the ability of any single entity to manage safety implications effectively.
What challenges do these companies face in establishing the standards body?
The companies face challenges due to their low ratings on the 2026 Future of Life Institute AI Safety Index, which assesses transparency and responsible development practices, as well as the lack of a formal organization as of mid-September 2026.
How might the establishment of these standards impact smaller AI firms?
If the major players succeed in creating these standards, compliance costs could disproportionately impact smaller AI firms, potentially leading to further consolidation in the industry.