By role

Who we help.

Different leaders answer for AI agents in different ways. Here is the job each one is doing, what gets in the way, and what good looks like.

Jobs to be done

The job, the risk, what good looks like.

CISO

Controls you can prove.

The job: Know which agents hold access, what it opens, and whether the controls around them hold.
The risk: Agents inherit broad access, some exceed their task, and agents built by business teams sit outside the usual review.
Good looks like: Every agent known, owned and limited to what it needs, with blast radius understood before an incident.

CRO

Agent risk inside appetite.

The job: Bring AI agents into enterprise risk management with owners, and report them against the appetite the board set.
The risk: Agents filed under “IT tools”, with no owner, no appetite limit and no view across vendors.
Good looks like: Agents on the risk register with owners, and trends the board can follow.

CIO

Scale agents with confidence.

The job: Say yes to more AI agents without losing track of what they are, who owns them and what they depend on.
The risk: Orphaned agents, vendors that change the model underneath you, and approvals that wait in the same queue regardless of risk.
Good looks like: An inventory you can stand behind, and approvals backed by evidence.

Compliance and internal audit

Evidence you can rely on.

The job: Answer auditor, regulator and customer questions about AI agents with evidence, not narrative.
The risk: Questions arrive before the evidence exists, and management’s own word is all there is to go on.
Good looks like: Independent evidence, mapped to NIST AI RMF and ISO/IEC 42001, with the method and date behind every finding.

Boards and audit committees

Oversight on record.

The job: Oversee AI agent risk without becoming technical experts.
The risk: Finding out after the fact, or receiving a dashboard nobody explains.
Good looks like: A short, plain answer each quarter: what we run, who owns it, what could go wrong and how we know.

Where we focus

Organizations that have to show their work.

Regulated enterprises

For example, financial services and healthcare: sectors where boards, auditors and regulators already ask how technology risk is governed.

Higher education and research

Campuses where research groups and departments build and adopt agents faster than central IT can review them.

FFRDCs and federal contractors

Programs that must show sponsors and contracting officers who approved each agent and what it can reach, across several model vendors.

Adopting agents

Risks for executives.

AI agents can do real work. Adopting them well means weighing a few ordinary risks up front.

Data exposure

Agents read and move data with whatever access they hold. Know which data each agent can reach, not just which data it was meant to use.

Over-privilege

Agents usually inherit a person’s or a service’s access, which is often broader than the task. Least privilege still applies; the actor is new.

Accountability gaps

For every agent, someone should be able to say who approved it, who owns it and whose identity it acts on.

Vendor and third-party agents

Agents built into vendor products act inside your environment too. Ask what they can reach, on whose credentials, and what happens if the vendor changes course.

Board and audit readiness

Boards ask whether agent risk is within appetite. Auditors ask about the access underneath. Being ready means having both before the question arrives.

Where to start

An honest inventory of agents, their owners and what each one can reach. Most of the rest builds on that.

Compare notes.

We’re shaping our first engagements with CISOs, CROs, CIOs, compliance and internal audit leaders. We’d like to hear how your organization approaches AI agents.

Compare notes