AI safety alarms senior Labour figures

Estimated reading time: 5 minutes · Last updated: 17 October 2026

Senior figures around the new Labour government are treating AI safety as a top security and political priority, worried both about immediate misuse and about falling behind public concern. As first reported by The Guardian, one cabinet minister said they feared concentrated, super‑intelligent capabilities landing in the hands of bad actors, rather than 'killer robots'. That alarm has been echoed by Kanishka Narayan, who called a state hack by rogue agents 'likely, if not inevitable', and by MI5 director Ken McCallum, who named extremist groups and AI‑enabled bioweapons among the biggest near‑term threats.

I have this sense that, within a few months, AI is going to be all anyone is talking about,

one cabinet minister

Why senior Labour figures are jittery about AI

Several cabinet ministers and senior Labour MPs now place AI safety alongside the cost‑of‑living and economic pressures on their priority list. They describe a political risk from failing to address real harms while also facing public anxiety about the technology’s speed and scale. One cabinet minister summed that unease in a direct image of concentrated machine intelligence ending up with the wrong actors, a concern framed as an ethical and security problem rather than a science‑fiction scenario.

Ed Miliband used his conference speech to warn of 'daunting' dangers and argued profits for tech companies must not come at the expense of children or humanity. Other ministers have expressed anxiety that Whitehall’s posture lagged public concern and needs to catch up, creating a policy tension between rapid adoption of productive tools and strong guardrails around harmful uses.

Security services highlight cyber and biological risks

MI5 director Ken McCallum told ministers that AI is amplifying the contest between attackers and defenders, with extremist groups singled out as likely early adopters of harmful capabilities. McCallum specifically flagged the risk of AI making design of bio‑harm easier to attempt, elevating concern within security circles about how model outputs could be repurposed for real‑world violence.

Geoffrey Irving, formerly AISI’s chief scientist, told MPs the open‑source ecosystem narrows the gap with commercial models quickly, typically catching up within less than a year, and that freely available models can already be adapted for disruptive ends. Experts retained a division: some stress existential scenarios; others, including Prof Elena Simperl of King’s College London, urge focus on immediate, evidence‑backed harms such as hacking and addictive design.

Industry testing, US ties and a regulatory moment

The business, innovation, science and trade committee held evidence with Anthropic, OpenAI, DeepMind and Meta about model testing and access. Committee chair Liam Byrne argued the new government’s stance differs from the previous administration’s closeness to big tech and called for the UK to seize the moment to lead on regulation, especially if US congressional dynamics change after the midterms.

The hearing revealed disagreement over whether the government should require testing of new models. The committee said there had been friction with the US and cited claims that the previous US administration asked firms not to share their latest models with the UK's AI Security Institute; ministers said cooperation has been retained. That tension frames a choice between model access and safeguards.

Defence research and the risk of being left behind

Defence sources raised concern that cancelling last year’s Dstl AI programme at Porton Down was shortsighted given the need for cutting‑edge research to harden the nation’s defences. The Ministry of Defence responded by pointing to a planned £400m investment in AI‑focused capabilities over the next four years, and said AI must be integrated across defence workstreams rather than run as a standalone project.

Former cabinet secretary Gus O’Donnell warned ministers to weigh the risk of being left behind against safety guardrails: accelerate capability where it is genuinely productive, but avoid errors that damage public trust. That balance — build capacity while proving safeguards — is the operational problem Whitehall faces as it moves from warning to delivery.

How the next phase could play out

The bottom line

Whitehall’s shift in tone — from cautious commercial friendliness to a security‑aware posture — has created a narrow window for policy action. The government must translate warnings from MI5 and committee hearings into practical controls and research capacity: mandatory testing or clearer model‑access rules, investment that restores Dstl‑level work, and international coordination through forums such as the G20. The immediate choices will determine whether the UK reduces real harms now and preserves the ability to develop or adopt beneficial AI in public services without ceding oversight to other powers.

What to watch

• Watch for the UK government’s AI agenda at its upcoming G20 presidency; no date has been set.
• Watch for the business, innovation, science and trade committee’s formal recommendations after it completes its evidence sessions; no date has been set.
• Watch for MoD detail on how the announced £400m AI investment will be allocated and timed; no date has been set.