ESEN
Assessment
← Research
Delegation framework

The AI-enabled vCISO

What gets delegated to an agent, what gets verified and what cannot be delegated. A four-level ladder and a measurement protocol, with explicit sources and limits.

Ricardo Martínez · 15 Oct 2026 · Editorial summary v1

Read PDF · EN ↗View materials ↗

What the agent makes cheaper is preparation, not accountability. The new risk is false assurance: a polished report that asserts compliance with no evidence behind it.

Demand for security skills is measurable, but we found no independent measurement of agent accuracy at producing compliance evidence. Three tests (reversibility, accountability and verifiability) set each activity's level, from automate to never delegate. No evidence, no claim: the gate blocks on any failure. 30-second animation, no audio.
Key data

What is measured and what is not

What it contributes

The delegation ladder sorts a vCISO's activities into four levels using three tests that do not depend on the model. On top of that come six design principles (among them linked evidence, a gate before delivery, the agent as a least-privilege non-human identity and a named human who signs), the likely failure modes and six questions to ask a provider in 2027.

Method and limits

Review of public sources with a cutoff of October 15, 2026. Only the ISC2 study was read at the source; the EU AI Act dates and the Mexican data protection law details come from third-party summaries and need verification. Vendor figures are treated as an indication. The ladder, the principles and the protocol are our own analysis, with no field validation and no figures of our own.

Keep reading

← All research