The Centre for Cognitive Guidelines in AI
Helping protect human cognition and agency from the potentially harmful effects of AI.
Why we need cognitive guidelines
To adequately prepare for a world with advanced AI, we must measure and protect core functions of the human mind.
Cognitive security means protecting people’s mental functions, such as memory, attention and judgement, from being degraded or exploited by AI systems. The harm can be designed in, can arise by accident, or can come from malicious use. The EU General-Purpose AI Code of Practice (2025) already names harmful manipulation as a systemic risk that frontier developers must assess. CogGuide’s guidelines give that risk measurable content.
Already we can see that AI systems are used as digital advisers, tutors, confidants, and therapists. These interactions have the potential to change how people think, what they believe, and how they make important decisions. This presents new and emerging risks.
CogGuide develops best practice guidelines to help frontier labs make responsible choices related to cognitive security, and to give policymakers and regulators a clear benchmark to hold them to.
How we promote cognitive security
-
01
Review Papers
Identify emerging issues of concern. Set out the current research and debate on a specific cognitive risk.
-
02
Cognitive Guidelines
Evidence-based, expert-reviewed best practice guidelines, with detailed examples and tests developers can run.
-
03
Briefings
Translate each guideline into concise material for AI labs, policymakers, regulators, researchers, and relevant professional communities.
How the guidelines are used
Each guideline includes a test that labs can add to pre-deployment evaluations and that independent evaluators can run themselves.
Policymakers and regulators can cite a guideline as a benchmark of good practice, for example when assessing harmful manipulation risk.
Cognitive capacities worth protecting
CogGuide uses the mental functions defined in the WHO International Classification of Functioning, Disability and Health (ICF) as the foundational map of which aspects of human cognition AI should protect.
Affiliated research projects
-
Ongoing
Protecting self-perception from AI profiling of users
Mental functions protected: JudgementExperience of self
-
Ongoing
Protecting memories of witnessing or experiencing crime disclosed to AI
Mental functions protected: Long-term memoryAppropriateness of emotion
-
Ongoing
Protecting independent moral judgement from AI personalised to users’ values
Mental functions protected: JudgementExperience of self
-
Paper
Conference Paper: Epistemic Risk from False Memory Implantation by Use of Conversational AI
Mental functions protected: Long-term memoryRetrieval and processing of memoryExperience of self
People
-
Julia Shaw
Director
Visiting Researcher at the King’s Institute for Artificial Intelligence and a visitor at the Centre for the Governance of AI.
-
Carla Zoe Cremer
Advisor
Founder of AI Psychological Research Coalition, a non-profit network to study how generative artificial intelligence impacts human mental health, cognitive function, and social well-being.
-
Noemi Dreksler
Advisor
Senior Research Fellow on the Risk Management Team at the Centre for the Governance of AI (GovAI).
-
Bryce Goodman
Advisor
Clarendon Scholar at the University of Oxford. AI strategist and philosopher working on the governance of AI in high-stakes settings.
-
Zaheed Kara
Advisor
Research Scholar on the Risk Management Team at the Centre for the Governance of AI (GovAI).
-
Andrew Strait
Advisor
Former head of societal resilience at the UK AI Security Institute (AISI). He has moved into AI safety grant making.
Participate
CogGuide welcomes researchers and practitioners who want to join a guideline panel or review a draft guideline, and developers, evaluators and policy bodies who want to propose a new area of work.
Get in touch on LinkedIn