Pioneering risk management where humans meet AI.

AI safety risks change established infrastructure industry risks.

We research where sociotechnical risk emerges and how. Build the tools to manage it, and help you evidence it to customers, investors, insurers and regulators. So you can go further, safely, with AI.

Book a call Built on research

“I use AI every day for safety management system legal tests and fact checking.”

A health and safety manager describing their use of AI. From primary industry research by our founding team.

Wicked problems...

Assurance is only as good as the risk management beneath it. Risk management is only as good as its grasp of relationships between data, humans, AI technology, processes and the real world. AI is changing these relationships and new risks are rapidly emerging.

Working from first principles we focus where sociotechnical risks impact safety, resilience of business critical decisions in infrastructure industries.

Widespread use

79%

Are using AI in critical physical infrastructure sectors, with over half for safety documentation. This could be positive, it could be creating risk - do you know which?

Your people are deskilling

Unmeasured

The more judgement your team hands to GenAI and agentic systems, the less they practise it. While the work gets more complex and the workforce gets more junior, thinner and busier.

You cannot see

72%

AI use in critical infrastructure happens on personal accounts, outside any system you control or can audit.

Our mission

Build resilience in societally critical infrastructure industries by mitigating risks where humans meet AI and real world risk.

What we do

Understand the risk. Then act.

Safety and business critical decisions are increasingly made by people and AI together. We study how the systems can go wrong, find the causes, and build systems to prevent it.

Learn

Testbed

For you if you want to understand how your AI system behaves, or will, in the real world.

Our testbed shows you how your AI system behaves before letting it loose in the real world.

We build a test version of your system. Run it under real or simulated conditions. Assess what happens against a defined risk framework. Our scientific method, applied to the whole system rather than the model alone.

Three ways to work with us:

Risk assessment. Identify where risks sit, and why.
Risk assessment and mitigations, evidenced. One pass, changes made and documented to the standard your stakeholders expect.
Bespoke testing projects. Designed for systems or questions that don't fit either.

AI system risk assessment

Test protocolBaselineStressObserveScore

AI technologies

KnowledgeStructure ▼

HumanRisk: deskilling

Physical asset

Process

Output

Critical asset context

Risk gauges

Deskilling%

Threshold exceeded

Over-reliance%

Opacity%

Knowledge structure▼ −5% %

Process safety▼ −6% %

Resilience▼ −1% %

Financial▼ −8% %

Build

AI systems for competent humans

For you if you need AI systems that build technical competence in users.

When we say AI system we capture the whole system that contributes to cognition: the technology, the knowledge it draws on, the people using it, the processes around it, and the assets it affects in the real world.

We design systems around the people who work with them and how they need to act on the output. Deliberately selecting human or artificial intelligence for the work at hand and responding to different user needs.

AI safety features are inherently designed in, for example enabling correct cognitive loading. Risk identification, analysis and mitigation is built in from the outset.

Off the shelf safe systems:

A coach — for safety, resilience and business critical domains.
Hybrid intelligence workflows — deliberate design of what people and AI do. AI inherently designed to build technical competence.

We build bespoke safe systems for high risk use cases. Typically following testbed assessment.

Manage

The platform to manage AI safety risks in critical infrastructure.

Our platform keeps the risk in view as your people, AI and assets keep evolving. Measuring sociotechnical risks and the impact on business safety, resilience, and critical decisions.

Risk controls are based on frontier research and scientific method. Leading to mitigations driven by the combination of leading insights and your use case data.

Mitigations building user competence. Each user contributing to the sharpening of user decision making, risk management capability and evidence.

Layer 3Assurance & reporting
Layer 2Critical CognitionAI safety risk management platform
Layer 1AI systemHuman · Knowledge · Process · Physical asset · Output
Real worldSocietally critical infrastructureOil & gas · Power · Gas · Rail · Roads · Water · Ports

Risk management from modelling, monitoring and assessment to mitigation and reporting.

Research. Rigour. Human.

Frontier research

Critical Cognition was started based on the scale of the challenges identified by our founders' work with Discovering Safety on two AI safety projects. Continued research at the intersection of AI safety with process safety, resilience and business critical domains is at the centre of our products.

Innovation for humans

Sociotechnical risks and challenges require a human centred response. Our innovation approach puts the understanding of human risks, such as deskilling from inappropriate transfer of cognitive tasks to AI, and their threats to decision making competence, at the heart of our products.

Start with one system

Bring us one AI system in safety or business critical work. We will work with you to understand and manage the risk.

Book a call