What AIdentity means in practice at Thiligence

1. Native identity for the AI workforce

An identity system built only around employee sign-in is no longer enough. Software agents can plan, call tools and act for people. A robot or drone can sense its surroundings and make decisions far from a browser. These actors need an identity and a clear answer to who owns them, what they may do and when their authority should end.

AIdentity brings human and non-human identities into one governance model without treating them as interchangeable. A person authenticates differently from a workload or a device. In AIrspace, the chain is explicit: an agent definition and version, the running instance with its own runtime identity inside an isolated microVM, the human sponsor behind the launch, and the host carrying that run. These remain distinct, connected records, so an investigation can follow who launched what and where it ran.

2. Your intent, in your language

Identity management should not require every administrator to become an expert in every protocol or remember where each setting lives. Someone should be able to ask, ‘Why does this person have access?’, ‘Which agents belong to this team?’ or ‘Help me review this role.’ The product should turn that intent into a clear, permitted operation.

Ask AIdentity provides that natural-language way into the product. It works within the signed-in person’s authority and can explain or help carry out supported tasks. It does not turn a conversation into permission. A change still meets the same authorization and approval controls as the corresponding operation elsewhere in the product. Ask AIdentity is working today and is continuously refined.

3. Identity guardians, working around the clock

An IAM system should help its administrators notice what changed, rather than wait for someone to open a dashboard. AIdentity has built-in specialist agents and analytics that work from identity activity and scheduled checks: detecting unusual behavior, assessing risk and preparing access recommendations.

The useful automation is a complete journey: observe a signal, gather relevant evidence, explain a recommended response, obtain the required decision and carry out the approved change. Around-the-clock monitoring depends on the configured services, sources and operating environment; it is not a promise that every possible threat is visible.

Here is the boundary that matters. Automatic analysis is different from automatic permission to change access. In the current production apply path, these proposed identity changes—including those classified as low-risk—are applied only after an eligible person approves them. Once an authorized person approves, the system carries out the checked action and records the result. A recommendation is never reported as an applied fix.

4. Graph intelligence that reveals hidden risk

A list of users and their direct roles tells only part of the story. A person can reach a sensitive resource through nested groups, inherited roles or a chain of delegated authority. Several individually ordinary relationships can combine into an uncomfortable concentration of power.

Identity Graph brings those connections into view. It helps teams investigate indirect access, potential attack paths and the reach of a compromise, with evidence behind the relationships. The question becomes more useful than ‘Does this identity have a role?’ It becomes ‘How did this identity get here, what could it reach, and why?’

Graph intelligence informs an investigation and a reviewed response. The currently shipped graph runtime is advisory; the graph does not independently grant access or silently enforce a remediation. That separation keeps useful intelligence from becoming an unreviewed authority.

5. Connect AI agents without expanding their authority

People increasingly want to work through an assistant they already know. AIdentity’s MCP interface gives compatible AI agents a governed way to use identity capabilities. An assistant can help a person investigate and request changes without making the person learn a second interface.

AIdentity’s MCP broker re-authorizes every tool call against the connected person’s current permissions and tenant. If someone may inspect a role but may not change its membership, their assistant may inspect it too; asking the assistant to add a member does not make that operation permitted. Connecting an assistant does not turn it into an administrator, and a sensitive write still follows its approval controls.

6. Human authority with enterprise accountability

The more capable automation becomes, the more important its boundaries become. RBAC, ABAC and ReBAC express who can do what and under which conditions. Privileged identity management limits when powerful roles are active. Privileged access management governs credentials and access to sensitive targets.

Approval and separation of duties keep consequential decisions with eligible people. Audit evidence links the proposal, the decision, the actor and the outcome. A system that can explain why it recommended a change—and show exactly what happened afterward—is more valuable than one that simply says its AI has taken care of it.

The product principle is practical: increase what people and agents can accomplish, while preserving who has the authority to decide.

Proof proportional to the action

Rolling out · Configuration and preview

Permission answers whether someone may act. Proof answers whether the person making this particular request has demonstrated enough assurance for what is at stake. Reading a routine setting and changing the rules that protect privileged access should not ask the same question of the user.

We are rolling out configurable action tiers. Once available in their deployment, an administrator can configure and preview the proof required for a sensitive operation: for example, a fresh phishing-resistant check with a passkey or security key for every high-impact action, or a recent second factor for a lower-risk operation. Approval of a tier change has a separate safeguard already implemented: the approving person must provide verified step-up evidence. That approval guard does not wait for the new per-operation enforcement cutover.

Configuration and previews are implemented, with customer availability following the deployment rollout. Live enforcement of the new action-tier requirements remains off in this rollout. The deployment operator must complete a separate enforcement cutover for each tenant before those requirements take effect. Existing authentication and approval controls continue to apply in the meantime.

What this looks like in a working day

Imagine that a change in identity activity surfaces an access pattern worth investigating. A specialist analysis brings the relevant context together and prepares a recommendation. An administrator can ask why it matters, inspect the evidence and decide what should happen.

If the response changes access, the governance path checks the required authority and approval. The action is then applied through the appropriate controls, and its outcome is recorded. The administrator does not have to repeat every approved step by hand, and the agent does not get to approve its own proposal.

The identity story reaches the edge

NHI—non-human identity—makes the definition larger than software running in a data center. A robot, drone, connected machine or other edge device needs its own identity, owner, credentials and lifecycle. The proof it presents and the permissions it receives should fit what it is and what it is doing.

AIdentity includes device-onboarding paths for FIDO Device Onboard (FDO) and Enrollment over Secure Transport (EST), alongside attestation checks that establish evidence about a device before it becomes active. That connects a device’s arrival to its owner, operational credentials and lifecycle, rather than leaving it as an anonymous endpoint.

A kill switch can suspend an agent or device identity and trigger the connected enforcement paths to restrict its operation. Enforcement depends on those integrations and connectivity; an identity control does not promise an instantaneous physical stop on disconnected hardware.

AIrspace adds governed execution for software agents in isolated microVMs. AIsense, which is under development, extends our direction toward secure model delivery and inference on intelligent edge devices. Device identity and edge inference are connected ideas, but they are different product capabilities.

Where we are taking AIdentity next

Directions · Under Development

The next steps are about giving people better evidence, safer automation and a broader reach.

• Change-impact previews

  Help a reviewer understand who and what a proposed identity change would affect before it takes effect.

• Broader, verified remediation

  Extend policy-bounded actions across connected systems, with clearer outcomes and recovery paths. More autonomy should follow stronger controls, not weaker approvals.

• Governed intelligence on devices

  Carry identity, approved AI capabilities and accountable operation into the edge through AIsense.

Bring a real identity question

For customers, the test is practical: can the system help your team understand and govern a difficult identity scenario? For investors, the opportunity is the same change in the enterprise: more actors can now take action, and all of them need accountable authority.

Bring us a hard access question, an agent workflow or an intelligent-device use case. We will show how AIdentity, AIrspace and the AIsense direction fit together—and be clear about what can be demonstrated today.