Question to the Department for Science, Innovation & Technology:

To ask the Secretary of State for Science, Innovation and Technology, what mechanisms exist for the AI Security Institute to receive systematic information about incidents involving autonomous or adaptive machine learning systems in critical national infrastructure as part of its intelligence capacity to research the development of AI capabilities that could contribute towards AI's ability to evade human control, as well the propensity of models to engage in misaligned actions.

The AI Security Institute (AISI) collaborates with leading AI developers to measure the capabilities of advanced AI and recommend risk mitigations, to ensure we stay ahead of AI impacts.

This close collaboration with industry enables information-sharing to mitigate risks. AISI’s testing has identified a large number of AI model vulnerabilities that labs (such as OpenAI and Anthropic) have addressed prior to release.

AISI is  researching  the development of AI capabilities that could contribute towards AI’s ability to evade human control, as well the propensity of models to engage in misaligned actions. AISI shares its insights with government departments to help manage the risks AI could pose to critical national infrastructure.

Through the Alignment Project – a funding consortium distributing up to £27m for research projects – AISI is supporting further foundational research into methods to develop AI systems that operate according to our goals, without unintended or harmful behaviours.