- oiste.org
- AI and Identity
- Identity for AI agents
AI and Identity / Post-quantum
Identity for AI agents
On 25 September 2026 OISTE and WISeKey extended the Quantum Root Key programme into a post-quantum root of trust for AI models, autonomous agents, connected devices and the humans who operate them.
Why identity first
Before an AI agent can access infrastructure, communicate with another AI or execute a transaction, we need to know cryptographically who that agent is, who authorized it and whether it has been modified.
Carlos Moreira, 25 September 2026
AI is moving from producing information to taking autonomous actions. In that setting the first question is no longer what a system said, but who acted, on whose authority, and whether the software that acted is the software that was approved.
The architecture answers those questions with certificates issued under the post-quantum root, keys held in hardware security modules, TPMs and secure elements, and signatures that relying parties can check without trusting the agent itself.
It is the enforcement layer of HUMAN-AI-T.
Capabilities
Seven things the architecture makes verifiable
AI model identity
A system can verify which model it is dealing with, and that an authorised organisation issued its identity.
Model integrity and provenance
Signatures and hashes show that model binaries, weights and components are unchanged since authorised release.
AI agent authentication
An autonomous agent authenticates before it reaches systems, APIs, financial services, IoT infrastructure or other agents.
Human-to-AI authorisation
Operators use trusted certificates to authorise an agent to act, and to modify, deploy or terminate it.
AI-to-AI trust
One agent verifies another’s identity and authority before exchanging information or executing a transaction.
Signed AI actions
High-impact actions are signed, recording which agent acted, under which identity and on whose authority.
Trusted audit trails
Signed events in tamper-evident logs give traceability across the AI lifecycle.
Where it applies
Sectors where an unverified agent is unacceptable
Financial services, healthcare, telecommunications, defence, energy, industrial systems, satellites, smart cities, connected vehicles and IoT infrastructure.