Hassabis Calls for US Frontier AI Standards Body

News

1. IBM Says It Faltered as AI Spending Upends Enterprise Buying

2. TCS to Power JFK Airport’s $9.5 Billion New Terminal

3. AI Use Rises but Engineering Depth Lags, Nasscom Study Shows

4. AI Agents Expose Insurance Blind Spots, Report Says

5. Birla Buys Sprng From Shell in $1.8 Billion Clean Energy Deal

6. India Uses Open-Source AI to Hunt Cyber Flaws

Google DeepMind Chief Executive Officer Demis Hassabis has called for a US-led standards body to evaluate frontier artificial intelligence models before release, warning that artificial general intelligence (AGI) may be only a few years away.

In a post published on Tuesday, July 14, Hassabis said AGI, which he defined as a system possessing the full range of human cognitive capabilities, was “probably only a few short years away.” He said society may later view the present period as the “foothills of the singularity.”

“AGI cannot be compared to standard technological breakthroughs, not even ones as consequential as the internet or mobile. It is much more akin to the discovery of electricity or fire,” he wrote.

Hassabis argued that AGI could accelerate drug discovery, clean-energy research and the development of advanced materials, while creating new cybersecurity, biological and potentially nuclear risks. Advances in frontier models were moving faster than researchers’ understanding of their behavior, he said.

His proposal would establish a Frontier AI Standards Body modeled on a federally supervised public-private organization such as the Financial Industry Regulatory Authority, which oversees US broker-dealers.

The body would be funded largely by the industry but overseen by a board including independent technical experts and representatives of the open-source community. It would work with federal agencies and US national laboratories to develop benchmarks and test models for dangerous capabilities.

Models exceeding specified capability thresholds would be classified as frontier models. Their developers would initially be encouraged to submit them for review up to 30 days before release.

Once the testing system proved reliable, Hassabis said clearance could become mandatory before frontier models were deployed in the US market. The framework would apply to qualifying models regardless of where they were developed or whether they were open or closed. Smaller models from startups and academic institutions would be exempt.

Assessments would examine capabilities involving cybersecurity, biological threats and other high-risk areas. They would also test whether agentic systems attempted to evade safeguards or displayed deceptive behavior.

Hassabis proposed updating the evaluations regularly, possibly every quarter at first. Although AI companies would initially help develop the benchmarks, the standards body would eventually create independent tests to reduce the risk that developers trained models specifically to pass them.

Frontier developers would also be encouraged to publish model cards, strengthen internal cybersecurity, vet key personnel and devote more resources to safety research.

The body could coordinate a slowdown among major laboratories if testing uncovered risks serious enough to justify one, Hassabis said.

In an interview with Axios, he said he wanted the organization operating before the end of 2026.

The proposal seeks to place technical standards between voluntary company safeguards and conventional government regulation. Its effectiveness would depend on whether competing developers accepted common tests and whether a US-based system could influence models developed outside the country.

Hassabis said the US framework could provide a starting point for international standards, although broader agreement would be needed because frontier AI risks would not be confined by national borders.

“The future is not yet written,” he wrote, urging governments and developers to use the period before AGI arrives to establish stronger safeguards.