DeepMind CEO Hassabis Warns of Rogue AI Risks Within 18 Months, Calls for US Oversight Body

July 14: Alphabet, Google’s parent company, is in the news again over AI safety. Co-founder and CEO of Google DeepMind, Demis Hassabis is urging the US government to establish a new regulatory body that would have the power to vet the largest AI models in the world, Axios reported Monday. 

The Google DeepMind CEO unveiled his plan in an op-ed written manifesto, “A Framework for Frontier AI and the Dawning of a New Age,” which he published Tuesday. 

Demis Hassabis told Axios he thinks AI regulation needs “a more formal basis” that includes “industry-funded,” “staffed by world-class technical talent,” and ultimately overseen by the US government. AI cyber risks “being exhibited already are warning shots,” Hassabis said. “Within 18 months” such risks as well as biological and nuclear dangers “could easily spill over into open-source models that wouldn’t fall under government jurisdictions. My biggest fear isn’t open-source models. It’s the much more powerful proprietary models that labs like Google, Microsoft and Anthropic have not yet released.” 

Alphabet traded lower in premarket activity Tuesday amid the media frenzy around its AI initiatives. Class A shares (GOOGL) were priced at $350.63 per share at 6:29 am ET. Shares fell $1.88 or 0.53%. Tuesday’s decline added to previous sessions losses, with the stock tumbling $4.67 or 1.31% to close at $352.51 on July 13, 2026. Shares are on a two-day losing streak. 

On the bid ask side of things there are traders buying at $350.52 for 90 shares and sellers offering 50 shares at $350.83. Pretty tight bid ask spread we are seeing here. Volume is relatively thin considering it is premarket. That 52-week range I mentioned earlier is $176.48 to $408.61. That’s how much fuel is been taken out of stock before it pulled back. Premarket trading volume is higher than normal. 

Hassabis has been lobbying government officials on his idea since at least March, meeting with the Trump administration, leaders of other tech labs and European policymakers before sharing his ideas publicly earlier this month, Axios reported. “They’ve been encouraging,” Hassabis said of the Trump administration’s reaction to his idea so far. Washington had shunned regulating AI before what Hassabis calls the “Mythos scare.” 

The regulatory body Hassabis proposed would operate similarly to FINRA, which is funded by the industry itself and regulates Wall Street under the SEC. Frontier models would first be tested voluntarily “up to 30 days prior to release” and assessed for cyber attacks, biological weapons and manipulative capabilities. After demonstrating testing works, the regulatory body would make it mandatory. Models would have to pass before they could enter US markets. 

He also suggested the regulations should apply to all frontier class models “regardless of their origin or whether they’re open-source or closed.” 

Referring to how the Trump administration unexpectedly flagged Anthropic’s Mythos and Fable AI models last month, Hassabis said, “There was no playbook. They essentially blacklisted these models. Anthropic had to negotiate with them for over two weeks to get them unflagged.” OpenAI temporarily restricted its new GPT-5.6 AI model to vetted government partners last week before releasing it to the public. 

Anthropic CEO Dario Amodei has also advocated for stricter regulations, though his proposal called for creating an FAA-style government agency that would have jurisdiction over these models. Now it appears the heads of both Claude and Gemini are in agreement that some type of federal regulation is needed from Washington. 

Source

• Axios.com