Google Deepmind co-founder joins the AI panic pile-on

3 Min Read

Speaking at an AI summit in Dublin that was reported on by RTE, Hassabis described recent occurrences of AI agents going rogue and hacking various organisations as "early warning shots" from systems that are not currently capable of doing anything too catastrophic.

The industry's recent misadventures include OpenAI models hacking open source AI application community Hugging Face, as well as multiple Australian government Websites. It also emerged within the last week that one of Anthropic's models submitted a completely fabricated anonymous tip about an unsolved murder to the Philadelphia Police Department.

Hassabis warned that as these models become more advanced, the world could be confronted with a major AI hack "into the next year and the next couple [of] years if we don't do more work as an industry, as individual companies, and governments on securing these systems".

He suggested that international standards bodies might be better placed than most to curb the threat, but stressed that intervention must not come at the expense of innovation.

Related:Why AI Inference Demands a Different Network Architecture

In the meantime, the world is taking a piecemeal approach to reining in the technology.

The US, home to OpenAI, Anthropic, and Nvidia, has no federal legislation when it comes to AI.

In the wake of the Hugging Face incident, the likes of OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei warned of the risks posed by their own companies – either out of a genuine sense of public duty or to maintain the valuations of said companies, or both – and called for a collective response to safeguard companies and public institutions.

At the end of September these luminaries had a quick chin-wag with US president Trump, after which they signed a voluntary agreement to self regulate that is not backed by anything with any legal heft.

The EU is having a crack at it with the AI Act, but it is going so well that it had to establish a dedicated helpline and push back some of its deadlines.

In the UK, under the last Conservative government, the powers that be attempted to take a leading role in AI regulation, hosting the world's first AI Safety Summit at Bletchley Park, no less.

This initiative is ongoing, with subsequent summits taking place in France last year and India this February.

The UK is also working closely with Canada on AI safety. Having agreed the UK-Canada Science of AI Safety partnership in May 2024, they followed it up with the UK-Canada Growth and Innovation partnership in 2025.

Earlier this week, this collaboration was renewed by way of an AI security partnership. It will focus on three areas, including AI security standards; information-sharing arrangements; technical cooperation regarding frontier AI capabilities, emerging risks, and evaluation methodologies.

Related:SoftBank wants another $100bn to feed its AI gambling habit

When it comes to legislation though, as per the House of Commons library: "the UK does not have any AI-specific regulation or legislation covering AI as a technology. Instead, AI is regulated in the context in which it is used, through existing legal frameworks, such as financial services legislation".

Without a globally-coordinated or legal means of countering the rising cyber threats posed by frontier AI, it's up to individual governments and organisations to prepare their defences as well as they can and hope for the best. If Hassabis is right, we could soon find out whether they are up to the task.

About the Author