Artificial intelligence
AI is in scope by two routes. The acute one is frontier capability: misuse for weapons or attacks, and the possibility of systems that humans cannot reliably control. The chronic one is structural: displacement of work at a scale that destabilises societies. The evidence for the two is very different, and we report them separately.
Why it is in scope
A product launch or a single company's layoffs do not qualify. A credible pathway does. For frontier risk that means a capability that could be used, or could act, at civilisational scale. For labour it means a sustained structural shift that could lead to mass immiseration and political breakdown. Much of this subject is projection. We try to keep clear what has been observed and what has been forecast.
Articles
-
Black October: AI finds the bugs, and the web scrambles
AI-assisted researchers found two critical flaws in PHP. The details leaked, governments took sites offline, and exploit attempts ran into the millions.
-
The backlash against the AI build-out
Hundreds of billions are going into AI data centres. Towns, counties and states are pushing back over power bills, water and land.
-
Frontier AI risk: what the evidence shows, and what it does not
Evaluation-aware models, safeguards added because tests could not rule out bioweapons help, and a first AI-run espionage campaign. Against that, no confirmed loss-of-control event.
-
Keeping AI out of the launch decision
Where AI and nuclear risk meet: human control over launch decisions, and the gap upstream of it.
-
The slow path: AI and the graduate jobs squeeze
Little sign of AI in the economy-wide numbers; a sharp squeeze at the entry level, and in junior software most of all.
Key facts
| Measure | Latest | Source |
|---|---|---|
| Expert survey: chance of extinction or permanent severe disempowerment from advanced AI | median 5%, mean about 16% (survey of 2,778 researchers, 2023) | AI Impacts |
| Frontier-AI safety law in force | California SB 53, from 1 January 2026 | Law firm and policy summaries |
| EU AI Act high-risk obligations | Deferred to December 2027 and August 2028 by the digital omnibus | Law firm summaries |
| Employment gap, ages 22–25, most- versus least-exposed jobs | 19% below (August 2026 update) | Stanford Digital Economy Lab |