Independent AI Verification
Approaches for independently examining AI claims, capabilities, performance, behaviour and evidence beyond provider assurance alone.
AI Verification Initiative
Advancing independent verification for the next era of AI.
The AI Verification Initiative advances the practical understanding, evidence, methods and collaboration organisations need to independently verify AI systems, infrastructure and ecosystems as they evolve.
Why the Initiative exists
AI capability is moving faster than independent evidence.
AI systems are becoming more capable, more embedded in organisational decision-making and more dependent on complex models, providers, infrastructure and workflows. Yet the evidence available to understand, assess and independently verify these systems is not developing at the same pace.
The Initiative exists to help advance practical understanding of what should be verified, what evidence may be required and how organisations can make more informed decisions about when and how to rely on AI.
The central question
How can organisations independently verify and responsibly rely on AI as its capabilities, behaviour, infrastructure and dependencies continue to evolve?
Priority areas
Six current priorities advancing the verification agenda.
These priorities provide an initial focus for the Initiative. They are not intended to define its permanent boundaries and will continue to evolve as AI capabilities, organisational needs and verification methods develop.
Initiative priority landscape
The Initiative's six current priorities provide an initial, evolving focus for developing independent AI verification as systems, infrastructure and operating environments change.
Evidence, Provenance and Accountability
How evidence is created, sourced, preserved, connected to decisions and used to support accountability across AI systems and workflows.
Agentic AI and Delegated Authority
How identity, authority, permissions, tools, delegation and accountability can be understood and verified where AI acts within or across organisational workflows.
Reliability, Behaviour and Runtime Assurance
How evidence, behaviour and trustworthiness can be assessed as AI systems, operating conditions and real-world use continue to change.
Sovereign AI and Infrastructure Dependency
How organisations can understand and verify ownership, control, jurisdiction, resilience, provider dependence and the infrastructure on which AI relies.
Standards, Assurance and Responsible Adoption
How standards, assurance practices, policy and evidence expectations can support responsible organisational adoption and more informed trust decisions.
How the Initiative advances the field
From shared questions to practical verification progress.
The Initiative connects organisational experience, technical research and wider ecosystem perspectives to identify priorities, explore useful approaches and advance practical verification across existing and emerging AI environments.
Think Tank Roundtables
Focused, closed-door discussions examining real verification questions, emerging risks and practical organisational needs.
Research and Briefings
Evidence-led research and briefings that clarify developments, identify evidence gaps and support a stronger shared understanding of AI verification.
Structured Working Sessions
Focused working sessions that help develop practical priorities, potential methods and questions requiring further exploration.
Ecosystem Engagement
Targeted engagement with policy, standards, research and industry communities where it can advance practical verification and responsible reliance on AI.
Participation
Participation is organised through the Just Verify Think Tank.
The Initiative defines and advances the wider verification agenda. The Think Tank provides the structured route through which organisations and contributors can participate and help develop practical priorities.
- The Initiativesets and advances the wider verification agenda.
- The Think Tankprovides the structured route for organisations and contributors to participate.
- Practical outputshelp inform research, methods and future verification priorities.
Get involved