OpenAI, Anthropic, and Google DeepMind Collaborate on Industry AI Safety Standards Body
The three major AI labs are actively working to establish an industry-led AI safety standards body, marking a concrete step towards coordinated development amidst growing concerns.
The story
A significant move towards industry self-governance emerged this week as OpenAI, Anthropic, and Google DeepMind confirmed their collaboration on establishing an AI safety standards body. These discussions, which have been ongoing since July, gained public confirmation on September 15.
The initiative builds on a proposal from Google DeepMind's Demis Hassabis for a Financial Industry Regulatory Authority (FINRA)-style body to test advanced AI models before public release. OpenAI stated its involvement on September 15, indicating a unified effort among the labs. This collaboration suggests a shift from public calls for caution to tangible steps in defining shared safety protocols, potentially influencing how future AI systems are developed and evaluated across the industry.
Who moved
OpenAI
What Changed: The company confirmed it is working with Anthropic and Google on creating an industry-led AI safety standards body.
Consequence: This move signals a concrete step towards collective governance and standardized safety protocols within the AI development landscape.
Anthropic
What Changed: Anthropic is actively participating in discussions with OpenAI and Google to form an AI safety standards body.
Consequence: This collaboration aligns with previous calls from its CEO for slower AI development and enhanced safety measures, moving towards shared industry oversight.
Google DeepMind
What Changed: The lab is collaborating with OpenAI and Anthropic on an AI standards body, building on an earlier proposal from its CEO, Demis Hassabis.
Consequence: Google DeepMind's involvement underscores a commitment to establishing pre-deployment testing and regulatory frameworks for powerful AI systems.
China's Foreign Ministry
What Changed: Spokesperson Guo Jiakun publicly rebuked Anthropic CEO Dario Amodei's call for continued US restrictions on advanced AI chips and chipmaking equipment.
Consequence: This response elevates a lab CEO's policy argument into a direct government-level dispute regarding access to frontier compute resources.
Market signals
Hyperscaler capital expenditure is projected to reach approximately $820 billion in 2026, exceeding their operating cash flow.
Implication: This indicates an unprecedented scale of investment in AI infrastructure by major cloud providers, often driven by demand from AI labs like OpenAI and Anthropic.
Anthropic's annualized revenue is approximately $65 billion, and the company has recently turned an operating profit, while OpenAI is projected to burn over $200 billion in cash through 2029.
Implication: These contrasting financial pictures highlight different business models and stages of monetization among major AI labs, with some showing strong revenue generation while others prioritize rapid, capital-intensive development.
Rogo Technologies announced $30 million in funding from investors including Citi, Barclays, and BNP Paribas for deploying AI agents in wealth management.
Implication: This signal indicates continued investor confidence in specialized AI applications for enterprise sectors, particularly in automating high-value tasks like portfolio reviews and client analysis.
Metacognition AI secured A$10 million in pre-seed funding to build a persistent-memory and reasoning layer for AI models.
Implication: This investment highlights a focus on foundational software layers designed to enhance AI model capabilities for complex, continuous operations, including robotics and industrial systems.
What we'll be watching
- OpenAI's annual developer conference, DevDay, on September 29, 2026, in San Francisco, with satellite events globally.
- The AI Societal Impact Lab Autumn 2026 Fellowship, a remote program for early-career researchers exploring AI's socio-political impact.
- AI Dev NYC, a technical conference in New York City for AI builders, hosted by DeepLearning.AI and Andrew Ng.
Reporting + analyst voices: grounded via Google Search at publish time.