Demis Hassabis wants a Wall Street-style referee for AI. He proposes a US-led Standards Body modeled on FINRA. It would vet frontier AI models before release and could enforce industry-wide slowdowns if risks rise.
Hassabis warns artificial general intelligence is “probably only a few short years away.” His idea is to catch dangerous capabilities early and coordinate labs to pause development if needed. The body would be funded mostly by AI labs, with a board including independent experts, Turing Award winners, and government officials.
Labs would submit their models up to 30 days before release. These “Frontier-class” models must pass safety tests before hitting the US market. The Standards Body would update benchmarks every quarter and apply rules to all models, except those from startups and academics. If risks grow, it could tighten restrictions and organize coordinated slowdowns among leading labs.
Hassabis briefed the Trump administration, rival labs, and European officials. He aims to have this body operational before the end of 2026. That timeline feels urgent given the current state of AI safety and governance.
Recent government actions show the stakes. Last month, the Trump administration froze Anthropic’s most powerful models overnight with an export order. OpenAI held back GPT-5.6 until it got government clearance, citing safety concerns. Yet, no clear public process exists for how government agencies evaluate or license these frontier models. The Department of Commerce’s Center for AI Standards and Innovation leads the effort but with little transparency.
Anthropic’s Mythos model was released to trusted organizations in April 2026. By June 12, it was blocked from foreign users due to security and jailbreak risks. The ban lifted on June 30. Anthropic’s Fable model also faced a brief foreign user ban for military concerns. Despite this, Anthropic technology has been used in US military operations in Venezuela and Iran.
Safety rankings paint a grim picture. Seven researchers and governance experts evaluated nine leading AI companies across six categories: risk assessment, current harms, safety frameworks, existential safety, governance, and information sharing. No company earned an A. Anthropic scored highest with a C+. Mistral, a French startup developing open models, ranked last.
Max Tegmark, MIT professor and president of the Future of Life Institute, expressed disappointment over Mistral’s last-place finish. “Europe has really… been a leader in AI safety,” he said. The overall assessment shows the global AI industry struggles to manage existential risks—especially those tied to human-level intelligence.
Many companies still develop open models that users can modify freely. Several, including Anthropic, reversed previous bans on military use of their AI technology. This dual use complicates safety and governance efforts. OpenAI’s president Greg Brockman and former White House AI advisor Sriram Krishnan have been involved in safety discussions, but the path forward remains unclear.
Hassabis calls current cyber risks “warning shots.” He warns graver dangers involving biological and nuclear tools could emerge within 18 months. His AI Standards Body aims to step in before it’s too late. If that sounds like a Wall Street referee with a power to hit pause, that’s exactly the point.
Based on
- Demis Hassabis wants a Wall Street-style referee for AI, with the power to hit pause — thenextweb.com
- A new plan emerges for AI apocalypse avoidance | Semafor — semafor.com
- How did the government decide OpenAI’s frontier model was safe to release? | TechCrunch — techcrunch.com
- AI industry fails to fight ‘existential’ threats, report says — ctvnews.ca
- Global AI industry falls short on safety, think tank warns — france24.com
