MI5 chief warns AI could 'change the stakes' on bioweapons

The head of MI5, Sir Ken McCallum, has warned that artificial intelligence could "change the stakes" on existing and even new bioweapons. He delivered the warning in his annual speech on threats to the nation, the traditional setting where the intelligence service's director general sets out the risks he believes Britain should be watching.

The stakes here are hard to overstate. When the leader of the UK's domestic intelligence service puts AI and biological weapons in the same sentence, he is framing a question that regulators, labs and security agencies are all now forced to answer. The Guardian reported the speech on October 14, 2026, noting that experts believe the technology could help terrorists and nation states with production — but that significant hurdles remain.

What did the MI5 chief actually say?

McCallum's core claim is contained in one phrase: AI has the potential to "change the stakes" on bioweapons. That is deliberately careful language from a career intelligence officer. It does not say AI has already enabled a new weapon. It says the technology could shift what is possible, both for weapons that already exist and for entirely new ones.

The distinction matters. Intelligence chiefs rarely speak in absolutes about threats that are still theoretical. Annual threat speeches are designed to flag risks while they are still on the horizon. McCallum chose to put AI-assisted bioweapons development in that category, alongside the established dangers his service tracks year to year.

The Guardian's framing, based on expert assessment, splits the problem in two. AI could lower barriers for bad actors — terrorists and nation states — specifically around production. At the same time, experts say significant hurdles remain before that potential becomes a working capability.

Who could benefit, and at which stage?

The source identifies two categories of actor: terrorists and nation states. That pairing is itself informative. Terrorism risk usually centers on capability — whether a group without state resources can acquire expertise and equipment it previously could not. Nation-state risk usually centers on speed and scale — whether AI accelerates programs that already exist.

The word experts used, according to the report, is "production." That is the stage where living agents must actually be made, stabilized and weaponized. It has historically been one of the hardest parts of any bioweapons program, and it is where AI assistance is now being discussed as a potential game-changer.

The counterweight is equally explicit: significant hurdles remain. Experts did not claim the barriers have fallen. They described a risk profile in flux, not a crisis that has arrived.

Why does this warning matter now?

An MI5 annual threat speech is not a research paper. It is a signal to government, industry and the public about where the security establishment believes attention belongs. McCallum's decision to foreground AI and bioweapons places the issue firmly on the UK's national security agenda.

The warning also lands in the middle of a live global debate over AI governance. Regulators have spent the past several years arguing about model safeguards, compute thresholds and evaluation regimes. A senior intelligence official now saying AI could "change the stakes" on bioweapons gives that debate a concrete, worst-case anchor. It connects abstract policy questions about model access and safety testing to a scenario security services genuinely worry about.

There is a second audience for this speech: the AI labs themselves. Companies building frontier models already face pressure to screen for biological-misuse capabilities before release. A public warning from MI5's director general raises the political cost of getting those evaluations wrong.

What remains unknown — and what comes next?

The source draws a clear boundary around what is established and what is not. Established: MI5's assessment that AI could shift the bioweapons calculus, and experts' view that production is the vulnerable stage. Not established: whether any actor has actually used AI to advance a bioweapons program, or how far the remaining hurdles would hold.

That gap between potential and capability is where this story will develop. McCallum framed the risk as one that could "change the stakes" — future conditional, not present tense. The next data points to watch are whether the UK government translates this warning into new policy on AI model access, and whether subsequent threat speeches escalate or soften the language.

For now, the assessment stands as a marker. The head of MI5 believes AI has put bioweapons risk in motion. The significant hurdles, in his experts' judgment, have not yet fallen — but they are the thing standing between a warning and an emergency.