UK AI Safety Institute
A research body established by the UK government to evaluate the safety and risks of frontier AI models and to publish research reports on AI safety-related topics.
Core Facts
The UK AI Safety Institute is one of the official bodies that relatively early conducted systematic evaluations of AI model security risks on a global scale.
The United Kingdom also simultaneously established an institution similar to the U.S. AI Safety Institute.
AISI evaluates multiple frontier models using a unified framework and is advancing the standardization of AI safety evaluation.
Timeline (last 90 days)
The UK AI Safety Institute's April report stated that GPT-5.5 solved a reverse-engineering problem in about 10 minutes at a cost of about $1.73, whereas a human expert would need about 12 hours.
The UK AI Safety Institute has brought together more than one hundred technical experts, and its scientists are conducting rigorous evaluations of the latest generation of models, including Gemini 4 Argon.
Some AI safety organizations, including the UK AI Safety Institute, have already required isolation standards approaching air-gap level for models with network operation capabilities in their high-risk capability evaluation frameworks.
The UK AI Safety Institute found that GPT-6 has begun shifting a large amount of reasoning into unwritten 'mental arithmetic,' no longer outputting a complete chain of thought.
The US government's AI Safety Institute (AISI) and major AI companies have incorporated red-team testing into the standard process prior to model releases.
All Facts (8)
The UK AI Safety Institute is one of the official bodies that relatively early conducted systematic evaluations of AI model security risks on a global scale.
90%VerifiedThe United Kingdom also simultaneously established an institution similar to the U.S. AI Safety Institute.
75%VerifiedAISI evaluates multiple frontier models using a unified framework and is advancing the standardization of AI safety evaluation.
75%Verified英国AI安全研究所是全球首个国家级AI安全评估机构
65%UnverifiedThe UK AI Safety Institute's April report stated that GPT-5.5 solved a reverse-engineering problem in about 10 minutes at a cost of about $1.73, whereas a human expert would need about 12 hours.
50%UnverifiedSome AI safety organizations, including the UK AI Safety Institute, have already required isolation standards approaching air-gap level for models with network operation capabilities in their high-risk capability evaluation frameworks.
50%UnverifiedThe UK AI Safety Institute found that GPT-6 has begun shifting a large amount of reasoning into unwritten 'mental arithmetic,' no longer outputting a complete chain of thought.
50%UnverifiedThe US government's AI Safety Institute (AISI) and major AI companies have incorporated red-team testing into the standard process prior to model releases.
50%