UK’s AI Safety Institute staff take leave for stress amid scrutiny
Burnout, restructuring, and the weight of testing frontier AI models are taking a toll on the government body tasked with keeping artificial intelligence safe.
The organization charged with evaluating whether cutting-edge AI models pose risks to national security is struggling with a very human problem: its own people are breaking down under the pressure.
Several staff at the UK’s AI Safety Institute, known as AISI, are currently on sick leave and receiving psychological counselling due to stress, according to a Financial Times report published on September 22, 2026. The causes range from grueling testing schedules to the existential weight of what they’re actually finding inside unreleased AI systems.
What’s driving the burnout
AISI’s core job is to evaluate frontier AI models before they reach the public. Staff on the cyber-security and bio-chemistry teams have raised alarms about AI’s growing ability to discover unknown software vulnerabilities and, more troublingly, to generate novel biological threats. One former employee described the resulting atmosphere inside the institute as stressful and, at times, toxic.
Tight deadlines compound the problem. Testing unreleased models from companies racing to ship products creates a constant time crunch.
A turbulent restructuring
In May 2026, the societal resilience team was merged into the human impacts unit. That reorganization slashed the combined headcount from roughly 15 researchers to just three.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
The fallout was swift. Andrew Strait, who headed the societal resilience team, resigned in July 2026.
The UK government has acknowledged the demanding nature of AISI’s work, pledging to ensure staff wellbeing and sustainable working conditions.
An institute still finding its footing
AISI was established in November 2023 as a government-backed body dedicated to evaluating AI risks, one of the signature outcomes of the UK’s AI Safety Summit at Bletchley Park. It was rebranded in February 2025 with a sharpened focus on national security and criminal misuse risks.
Despite the high-profile mandate, the institute operates with a notable limitation: it conducts voluntary evaluations of frontier AI models but lacks the statutory authority to compel companies to submit their systems for testing or to block a model’s release.