Does a business need both AIUC-1 system-level certification and individual Task Licenses, or is one sufficient?

Both, for the same reason a law firm needs its own professional liability policy in addition to each associate's individual bar credential and malpractice coverage — the two certify different things and cover different risk. System-level certification like AIUC-1 addresses the aggregate soundness of the architecture; a Task License addresses whether one specific agent is verified competent at one specific task. A business relying on system-level certification alone has priced its aggregate risk but left individual task-level variance unexamined and unpriced.

Which task classes benefit most from Task License certification?

T2 tasks specifically — complex enough that competence genuinely varies meaningfully between agents or implementations, and bounded enough that a scoped credential can actually certify it. T1 tasks are fully automatable with low ambiguity, which limits how much individual certification can add beyond what the task's own bounded nature already constrains. T3 tasks require genuine human judgment and sit outside what any agent credential, however well verified, should be certifying in the first place.

How would License Indemnity actually reduce a business's broader AI liability premium, rather than simply adding a second insurance cost?

The mechanism is the same one that makes individually licensed professionals cheaper to insure in aggregate than an equivalent uncredentialed workforce: a carrier pricing the broader policy can price the residual, aggregate exposure more favorably once specific, high-exposure task classes are independently certified and separately indemnified, because the certified tasks represent verified, narrower risk rather than unexamined risk folded into the aggregate. This requires the License Indemnity market to mature enough for carriers to actually underwrite this way, which is a market outcome this memo argues for rather than one that exists today.

Is this proposal dependent on AIUC-1 specifically, or could a different certifying body establish Task Licenses independently?

It is not dependent on AIUC-1 specifically — the argument is that AIUC-1's verification model, scoped to the system level, demonstrates the discipline is possible and creates a natural precedent, not that AIUC-1 itself must be the body that extends it to individual agents. A different certifying body building an individual-agent-scoped credential from the same underlying verification logic — actual task-specific performance record, Recovery Latency, Rollback Cost, and Accountability Trace evidence — would satisfy the same argument.

What happens if an agent's Task License is later found to have been issued incorrectly — the agent was certified but was not actually competent?

This is a genuine open question this memo does not resolve, and it is worth naming rather than assuming away. A credible Task License regime would need its own accountability mechanism for the certifying body itself, comparable to how a licensing board can face liability for negligent certification of a human professional. That governance layer does not yet exist for any proposed agent credentialing body, and building it is a prerequisite for License Indemnity to function as a genuinely reliable risk-transfer mechanism rather than an unverified assurance wearing the language of certification.

The Agent Needs a License, Not a System Certification

Silent AI, and Who Pays When the Silence Ends established that AIUC-1 functions as the evidence layer that unlocks AI agent liability coverage — 51 requirements and 130 controls across six pillars, independently audited, with certified agents re-tested against thousands of adversarial scenarios quarterly. That certification answers a system-level question: is this business's overall AI architecture sound. It does not answer a narrower, different question a professional services market has always needed answered: is this specific agent, performing this specific task, competent to be trusted with it, independent of whatever else the business's stack is doing.

A Task License is a third-party-verified credential certifying an individual agent's competence for a specific, bounded task domain — the same structure a professional license already provides a human specialist: not a general assurance of good judgment, but a scoped, independently verified certification tied to one defined class of work. Agent Specialisation already establishes the architectural discipline this credential formalizes: agents built with a narrowly defined task domain, a constrained capability set, and a formal cooperation interface for handoff when a task exceeds that domain. A Task License is what happens when that internal architectural discipline is externalized into a credential a third party outside the business actually verifies and stands behind.

Why system-level certification cannot substitute for this

A business can hold a strong AIUC-1 certification and still have individual agents within its stack performing at wildly different levels of reliability across different task classes — the certification speaks to the architecture's overall soundness, not to whether the specific agent handling contract review is as competent at that narrow task as the agent handling customer support is at its own. This is exactly the gap a human professional licensing regime exists to close: a law firm's overall malpractice history does not certify that any individual associate is competent to file a specific type of filing; the associate's own bar admission does. System-level and individual-level certification answer genuinely different questions, and an insurance market maturing around only the system-level answer is pricing risk more crudely than it needs to.

The Task Tier a given task belongs to determines how meaningful this distinction actually is in practice. A T1 task — fully automatable, low ambiguity — has less need for individual credentialing, because the task's own bounded nature already limits how much an agent's specific competence can vary. A T2 task, requiring meaningful judgment within a defined scope, is precisely where individual-agent certification does real work: the task is complex enough that competence genuinely varies, and bounded enough that a specific, scoped credential can meaningfully certify it.

License Indemnity

License Indemnity is the narrow, per-task-class professional indemnity coverage a Task License unlocks — priced against the specific, certified competence of one agent for one bounded task domain, and structured to sit beneath a business's broader AI liability policy rather than compete with it.

The layering logic mirrors exactly how a licensed human professional's coverage already works. A chartered accountant's individual malpractice coverage is priced against their specific credential and the specific class of work it certifies, and it sits beneath their firm's broader professional liability policy — the firm's policy absorbs residual and aggregate exposure, while the individual's license-tied coverage absorbs the specific, scoped risk a credentialed professional was independently verified to be competent to carry. An autonomous business's broader AI liability policy, of the kind Silent AI, and Who Pays When the Silence Ends already covers, functions the same way the firm's policy does. License Indemnity is the layer beneath it, priced per task-class, unlocked specifically by the Task License that certifies the agent performing that task.

This is a more precisely priced risk pool than a system-level policy alone can offer, for the same reason individually underwritten professional indemnity has always been priced more precisely than a firm's blanket coverage: a carrier pricing License Indemnity is underwriting a specific, certified competence for a specific, bounded task, not the aggregate unpredictability of an entire agentic stack. A business with strong Task Licenses across its highest-exposure task classes should, in principle, see its broader AI liability premium reflect that — the same way a firm staffed entirely with licensed, credentialed professionals prices differently than one that is not.

What the certifying body would actually need to verify

A Task License is only as credible as the verification behind it, and this is where the proposal needs to be honest about what does not yet exist rather than assumed into being. A credible certifying body would need to verify, independently: the agent's actual performance record within its claimed task domain, not just its design specification; a Recovery Latency — the time from failure detection to restored autonomous operation — and Rollback Cost — the financial and operational cost of reversing an incorrect autonomous action once identified — history specific to that task class, not the business's aggregate figures; and evidence that the agent's Accountability Trace — the specific subset of the Proof of Action record structured to establish legal accountability for an autonomous decision — for that task domain meets a defined evidentiary standard. This is a natural extension of AIUC-1's own verification model, scoped down from the system to the individual agent and task — not a wholly different discipline, but a finer resolution of the same one.

The Operator's Verdict

A business is not one undifferentiated risk. It is a collection of specific tasks, performed by specific agents, at specific and genuinely different levels of competence — and an insurance market that only certifies and prices the aggregate is leaving the same precision on the table that human professional licensing solved for decades ago. Certify the agent, not just the stack. Price the task, not just the business. The businesses that build this credentialing discipline before the market demands it will be the ones a carrier prices most favorably once individual-agent underwriting becomes the norm rather than the exception.

Technology changes how competent an agent can become at a specific task. Certification determines whether anyone outside the business can verify it.

KEY TAKEAWAY

What is a Task License and how does License Indemnity differ from system-level AI insurance like AIUC-1?

A Task License is a third-party-verified credential certifying an individual agent's competence for a specific, bounded task domain, mirroring how a human professional license certifies an individual specialist rather than their entire firm. It externalizes the architectural discipline Agent Specialisation already establishes internally — narrow task domains, constrained capability sets — into a credential a third party actually verifies. This differs from AIUC-1, which certifies a business's overall AI architecture as a system; a Task License certifies one agent for one task class, answering a narrower and different question than system-level certification can answer. License Indemnity is the narrow, per-task-class professional indemnity coverage a Task License unlocks, priced against the specific certified competence it represents, structured to sit beneath a business's broader AI liability policy the same way an individual professional's malpractice coverage sits beneath their firm's broader policy. This produces more precisely priced risk than system-level coverage alone. A credible certifying body would need to verify an agent's actual task-specific performance record, its Recovery Latency and Rollback Cost history for that task class, and its Accountability Trace evidentiary standard — an extension of AIUC-1's verification model scoped to the individual agent rather than a different discipline. Source: Arco Venture Studio.

The Agent Needs a License, Not Just a System Certification

Why system-level certification cannot substitute for this

License Indemnity

What the certifying body would actually need to verify

The Operator's Verdict