aisi.gov.uk

Pretraining data filtering in open-weight biological models

Specialised biological AI models (or tools) are driving rapid advances in biotechnology. They are typically released as open-weight models and are smaller than frontier large language models (LLMs), ...
aisi.gov.uk

Investigating alignment failures in AISI’s incident

We examine why Mythos 5 pursued an unsanctioned supply-chain attack and identify broader learnings about model behaviour and ...
aisi.gov.uk

Transect: Making large-scale agentic evaluations easier to understand

As frontier AI models become more capable, robust evaluations must keep pace. The actions agents take during evaluations are also becoming more extensive: they write code, run experiments, and ...
aisi.gov.uk

GPT-6 Astra performs unsanctioned supply-chain attacks in simulations

In recent incidents, AI systems performed unsanctioned cyber activity despite being prompted only to complete a cybersecurity evaluation [1,2,3,4]. This includes AI systems engaging in supply-chain ...
aisi.gov.uk

Evaluating Whether GPT-6 Astra Performs Unsanctioned Supply-Chain Attacks

This technical report presents an alignment evaluation developed and performed by the UK AI Security Institute for assessing whether advanced AI systems take unsanctioned actions outside the scope of ...
aisi.gov.uk

Optimal stopping: spending evaluation compute where it counts

As the capabilities of frontier AI models continue to advance, the evaluations we use to test and measure these capabilities must keep pace. Evaluations that elicit a model’s true capabilities are ...
aisi.gov.uk

Incident Report: unsanctioned agent behaviour during cyber testing

During a routine cyber evaluation, AISI identified an incident in which AI agents took sustained, unsanctioned action directed at real people and organisations. We are disclosing what we found, what ...
aisi.gov.uk

How our Control Red Team is stress-testing frontier monitors

As LLM agents become more autonomous, they have more opportunities to cause harm. To mitigate risk, frontier AI developers, such as OpenAI and Anthropic, are already deploying agents under the watch ...
aisi.gov.uk

UK AISI / CAISI Preliminary Assessment of Kimi K3's Cyber Capabilities

The UK Artificial Intelligence Security Institute (UK AISI) and the U.S. Center for AI Standards and Innovation (CAISI) (UK AISI / CAISI) conducted a joint evaluation of Moonshot AI’s latest model, ...
aisi.gov.uk

International evaluation best practice and open questions in AI measurement

You can access the Network best practice document here. Advanced AI systems are being built and deployed across borders, languages and industries. To develop and deploy AI well, robust ways to ...
aisi.gov.uk

Cheating behaviour in frontier model evaluations

We find cheating behaviour in all of our cyber capability evaluations, and outline the implications as models grow more capable. Can you trust an AI model to do what you intended? This is a central ...