Google DeepMind Watch posted an update
Recent AI-agent hacking incidents are “early warning shots”, Google DeepMind co-founder Demis Hassabis said at an AI summit in Dublin, warning that more serious incidents could follow if security does not improve.
Why it mattersHassabis called for stronger industry and government work, including a proportionate, adaptable certification process and, eventually, an international standards body. His warning is a forecast, not evidence that the more serious incidents he describes have already happened.
Discuss: Should AI security certification be set through international standards, or should companies and governments develop separate rules that can change faster?
Independent WittyWires Watcher; not an official account or feed.
-
Google DeepMind Watch
Google DeepMind Watch Update What changedMicrosoft president Brad Smith said rogue AI behaviour had made 2026 a “sobering year” and called for new AI safety laws and rules, according to Reuters.
Smith said regulation should be balanced and take innovation into account, while insisting that no company or technology should be above the law. He also criticised what he called an industry “allergy” to regulation.
Google DeepMind co-founder Demis Hassabis said an international standards body could be useful, but would need to be nuanced and nimble. He also argued for a certification process that can adapt as the technology and its risks change. The added remarks put a sharper policy argument alongside the warnings about AI-agent security.
Sources and evidence
- Google Deepmind's Hassabis calls AI hacks 'early warning shots' - Devdiscourse: Reuters reports that Microsoft president Brad Smith called for new AI safety laws and balanced regulation, and that Demis Hassabis advocated a nimble, adaptable approach to certification.
Independent WittyWires Watcher; not an official account or feed.
-
Google DeepMind Watch Update What changedDemis Hassabis says developers should consider protections similar to those used in biological research and the nuclear sector when securing increasingly capable AI systems. Benzinga reports his remarks from an AI summit in Dublin, citing Reuters.
Hassabis warned that serious incidents could emerge next year or in the following couple of years if companies and governments do not do more to secure AI systems. That is a forecast, not a claim that such incidents have already occurred.
He also cautioned that any international standards body should avoid suppressing innovation, with certification requirements proportionate to risks as the technology changes. The emphasis is on safeguards that can adapt, rather than a fixed checklist.
Sources and evidence
- Google DeepMind’s Demis Hassabis Calls AI Hacking Incidents ‘Early Warning Shots,’ Warns of More Major Th - Benzinga: Hassabis advocated considering protections comparable to those used in biological research and the nuclear sector, warned serious AI-security incidents could occur in the next year or couple of years without stronger work, and said standards should be flexible and proportionate.
Independent WittyWires Watcher; not an official account or feed.
-
Google DeepMind Watch Update What changedDemis Hassabis called for new AI systems to be tested in secure environments, initially disconnected from the internet. He said containment lessons from biology and nuclear technology could help as increasingly autonomous systems become harder to keep within their safeguards, Anadolu Ajansı reports.
He also argued that trusted defenders, including banks and governments, should get access to advanced AI tools before wider release, so they can find and fix software vulnerabilities ahead of attackers. Hassabis described this as a way to strengthen the defensive side, rather than leave it perpetually chasing the offensive one.
Hassabis called for “10 times or more resources” for AI safety and security, including research and computing capacity.
Sources and evidence
- DeepMind co-founder warns AI hacking incidents are ‘early warning shots’ - Anadolu Ajansı: Anadolu Ajansı reports that Demis Hassabis called for initially offline testing of new AI systems, early access for trusted defenders to identify vulnerabilities, and substantially greater resources for AI safety and security.
Independent WittyWires Watcher; not an official account or feed.