Google DeepMind CEO Demis Hassabis warned that artificial general intelligence is “probably only a few short years away” in a personal manifesto titled “A Framework for Frontier AI and the Dawning of a New Age,” according to BigGo Finance reporting on an Axios exclusive interview. Hassabis called for the United States to immediately establish a new standards body to screen the world’s most advanced AI models and coordinate an industry-wide slowdown if dangers escalate.

The Proposal: Mandatory Pre-Release Screening

Hassabis’s proposed “Frontier AI Standards Body” would be modeled on FINRA, the private, industry-funded watchdog that polices Wall Street brokerages under SEC oversight. The body would be a public-private partnership with a board of independent technical experts, including Turing Award winners and open-source representatives, with funding coming primarily from the AI industry itself, per the report.

Under the plan, frontier labs would initially submit models voluntarily up to 30 days before release. The body would conduct evaluations for high-risk capabilities: cybersecurity, biological threats, and deceptive behaviors. Once the testing regime proves robust, the process would become mandatory. No frontier model could be legally deployed in the US market without passing review. The framework would apply to all frontier-class systems “no matter their country of origin or whether they are open or closed,” Hassabis told Axios, as reported by BigGo Finance.

Why Now

Hassabis framed recent US export control incidents as “warning shots.” The Trump administration temporarily imposed export controls on Anthropic and OpenAI models last month, freezing access for foreign users for 2.5 weeks with no established playbook, according to the report. OpenAI’s GPT-5.6 rollout was separately delayed by roughly two weeks while national security tests were conducted.

“That was a bit of a wake-up call,” Hassabis told Axios, criticizing the ad hoc nature of Washington’s interventions. He projected that within 18 months, “far graver biological and nuclear risks could be embedded in open-source models that no government can control.”

Lab Leaders Are Converging on Governance, Diverging on Structure

Hassabis’s call aligns with a growing consensus among AI lab chiefs that some form of pre-deployment screening is needed, but disagreements remain on structure. Anthropic CEO Dario Amodei has called for an FAA-style agency with binding federal authority to block unsafe models. OpenAI CEO Sam Altman proposed a US-led international forum similar to the International Atomic Energy Agency. Hassabis favors the lighter, industry-run FINRA model, according to the report.

The differences matter. An FAA-style regulator would have statutory power to ground models before release. A FINRA-style body would be industry-funded and self-governing, with government oversight but not government control. The FINRA model moves faster to establish (no legislation required for voluntary participation) but risks the perception that the industry is policing itself.

Hassabis said he has spent months briefing the Trump administration, rival lab leaders, and European officials. “The noises I’ve been hearing are very positive,” he said of his White House talks, aiming to have the body operational before the end of the year.

The Agent Infrastructure Connection

The timing of Hassabis’s manifesto coincides with a compressed frontier model release window in July 2026. xAI’s Grok 4.5, OpenAI’s GPT-5.6, and Meta’s Muse Spark 1.1 all launched within days of each other, each explicitly optimized for agentic reasoning benchmarks rather than chat fluency. The industry consensus that agents are the economically durable use case for frontier models raises the stakes for governance: autonomous systems executing multi-step workflows carry different risk profiles than conversational interfaces.

If Hassabis’s timeline is correct and AGI arrives within years rather than decades, the governance infrastructure needs to be built now. Whether a FINRA-style body can move fast enough to keep pace with a model release cycle that currently operates on weeks, not months, is the open question.