Who Owns AI Risk When You Don't Have a CISO?

At most firms we talk to, AI arrived the way email and smartphones did: staff started using it before anyone decided they could. Now the questions are arriving. A client asks in a security questionnaire whether you use AI with their data. The cyber-insurance renewal asks about it. The auditor asks. And there's no obvious person to answer, because the firm has an IT manager, maybe an outsourced provider, and no CISO.

This post is for the owner, COO or IT lead at a 50–500-person firm who has realized AI risk currently belongs to nobody.

Why it falls between chairs

• IT sees a policy decision: which tools are allowed, and what data may go in.
• Compliance or legal sees a technical question: what the tools do with data, and what can be blocked.
• Finance sees an audit question: was AI output relied on in the close?
• Staff see a useful tool and carry on, often on personal accounts.

Regulators are noticing the same gap. Speaking at a June 2026 conference, as reported by Thomson Reuters, an SEC deputy chief accountant asked: "How much do we understand about AI that may be proliferating throughout our business and financial reporting processes?"

What "owning AI risk" actually means

COSO's February 2026 publication, Achieving Effective Internal Control Over Generative AI, puts it plainly: "Clear ownership of AI systems is essential." It asks for a named owner for each tool or platform, "responsible for its objectives, risk profile, and controls," and says "Owners must have both the authority to make necessary changes and the accountability for their outcomes." On oversight: "The board of directors must have visibility into GenAI use and associated risks."

That sentence about authority is why AI risk usually can't sit with a mid-level IT admin. Approving or blocking a tool the CFO's team depends on is a leadership decision.

A right-sized framework: NIST's four functions

The NIST AI Risk Management Framework was released January 26, 2023 and, in NIST's words, "is intended for voluntary use." NIST added a Generative AI Profile (NIST AI 600-1) in July 2024. The framework's core has four functions: GOVERN, MAP, MEASURE and MANAGE. For a mid-sized firm that mostly uses AI rather than building it, here's what each one looks like in practice:

That's a governance program a firm your size can actually run. It doesn't need a committee or a new platform.

The questions a CISO would ask first

1. What AI are we using today, and on whose accounts? If nobody can answer, start there. See what IT does first.
2. Which of our data must never go into AI, and can we detect it if it does? See mapping the AI policy to controls.
3. Do our contracts allow it? Client contracts, BAAs and NDAs may restrict where their data goes. For health data, the answer depends on the exact tier: see our ChatGPT and Claude breakdowns.
4. Is AI output relied on anywhere that matters? The close, client deliverables, HR decisions. For finance, see reliance in the close.
5. What will we tell the board, the insurer and our clients? One page, updated quarterly.

Where a virtual CISO fits

Most firms this size don't need a full-time CISO to answer those questions. They need someone senior to own them for long enough to set up the structure, then hand a working process to internal staff. We do both sides of this at Pro IT NW: security leadership and the AI tooling itself.

• Virtual CISO: a fixed-scope security-leadership engagement, not a retainer. AI governance can be scoped in: ownership, risk register, approved tools, board reporting.
• AI usage review: the hands-on part. Discovery of what's in use, a data-exposure check, the acceptable-use policy and approved-tool list, and setup of whichever tool your team prefers: Claude, ChatGPT, Copilot or another.

Either way, this is operational and technical security leadership, not legal advice.