Contents · 6 min read
01 / 06 · The gap
Why can’t the CIO make every AI use safe?
In short
Strong tech controls are vital, but they cannot carry every AI risk. Tight rules can push staff into personal tools that the firm cannot see. The CEO must own the company-wide trade-off within the board’s AI risk appetite.
Take a specialty insurer using AI to speed standard-risk quotes. Underwriters keep their judgment for complex risks. The CIO can bound the tool. The CEO must judge whether those bounds protect the firm without slowing useful work.
The firm is right to set limits. AI can read records and act in core systems. Those acts need firm control.
For the insurer, though, the safe route must be easy. If each task takes too long, staff will seek a faster path. Some may move work files into personal tools.
If they do, the firm loses sight of the data and the output. Controls meant to cut risk have moved it out of view.
This can happen when firms block useful tools without a good route in place. Staff may find workarounds with no company checks, as this piece on the AI leadership gap explains.
The CIO can run the controls. The board must choose how much AI risk the whole firm should take.
02 / 06 · Risk appetite
Who owns the company’s AI risk appetite?
In short
The board sets the risk line. The CEO weighs unsafe use against the harm from moving too slowly, then runs the company within that line. A change to the line goes back to the board.
The board owns how much AI risk the firm will carry. It states what return would justify taking that risk. This decision cannot be delegated, as the EU AI Act checklist sets out.
For the insurer, broad write access without approval can let an error spread through company records. Requiring approval for every routine quote indefinitely may slow the work. Both can hurt the firm.
The CIO can show what can be kept safe. The underwriting head can show what the rules do to the work. The CEO must weigh the effect on the whole company.
That includes the cost of delay. Faster rivals can win standard risks while the firm waits. Yet speed alone cannot set the risk line.
The CIO can then turn that line into technical bounds.
03 / 06 · Tech bounds
What technical boundaries belong to the CIO?
In short
The CIO or CTO owns how AI reads and acts across company systems. This includes access, approval rules and records of each action. Access should widen only when the firm has proof that it is safe.
IT owns agent identities and permissions. It also keeps approval and audit records.
That is the core of an AI-augmented IT department. Business teams use approved tools. IT sets the technical limits around them.
For the insurer, the quote tool should read only the records it needs. Its access should start narrow. A person should approve each write at first.
If people stop changing the proposed writes, the CIO has evidence that the tool can make those narrow writes without correction. The firm can then widen access one step at a time. This is why AI agent access to company systems is a series of choices.
The CIO should review denied actions and human overrides. The review should also cover incidents in which an agent played a part.
Those checks make the technical bounds real. Inside those bounds, each use still needs a business owner.
04 / 06 · Business owner
Who owns each business use?
In short
The leader who owns the work also owns the results of its AI use. That person must name who can stop the system. They must also name the exact point where that can happen.
For the insurer, the underwriting head owns the quoting use. The CIO controls the tool, but does not own the quote.
The tool can prepare the quantitative layer of standard quotes. That speeds routine work. Complex risks still need an underwriter’s judgment.
The named underwriter must be able to stop the quote before it leaves the firm. A vague promise of human review is too weak.
People who carry the result must make the key call. That is also the rule for an AI-augmented management team. AI can prepare the work, while people keep the choice.
If underwriters begin changing more of the proposed quotes, the owner reviews the use. When the risk changes, that owner raises it.
The CIO keeps the tool inside its bounds. The business owner remains answerable for its use. Hard cases then go to Legal for advice.
05 / 06 · Legal advice
What does Legal decide?
In short
Legal advises on hard cases and tells the firm which rules apply. It can challenge the business owner and test the evidence. The business owner still decides whether an allowed use is worth its risk.
Legal has a key role. It can find gaps in the facts. It can also explain which duties apply.
For the insurer, Legal may need to review a classification. It may ask for more proof before the use goes ahead. It may also challenge the owner’s view.
An AI-augmented legal and compliance team can prepare much of this work. People still approve departures and accept risk.
Legal cannot make the business call for the underwriting head. A use may be lawful and still be a bad bet. It may also carry risk that the firm has chosen to accept.
The named owner decides whether the gain is worth that risk. If the choice sits outside the board’s line, it moves up. That decision and its escalation route belong on the firm’s live AI-use map.
06 / 06 · The map
Map every live AI use to one owner chain
In short
Put every live AI use on one map. Record its business owner and human stop point. Add its technical bounds and Legal adviser. Any use outside the current risk line goes to the CEO. Only the board changes that line.
Start with the work being done. A tool name alone tells you too little. For the insurer, “AI helps underwriters prepare standard quotes” is a use.
Place that use against the board’s agreed risk line. Name the underwriting head as its business owner. Name the underwriter who can stop a quote before it leaves the firm.
Record that IT limits which records the tool can read. At first, a person approves each write. Mark Legal as the adviser on classifications.
The CEO can now see whether the use remains inside the agreed line. If the tool gains wider access or starts making a choice reserved for an underwriter, the CEO reviews it before the change goes ahead. If the change requires a new risk posture, the CEO takes it to the board.
After an incident, any use shown to sit outside the agreed risk line requires CEO review. If the incident shows that the risk posture must change, the CEO takes that decision to the board.
Keep the map live. Review each use when its access grows. Review it again when its business effect changes.
Start the next board review with any use that sits outside the line.
Set the access boundary
The management team now needs to turn the board’s risk line into a live owner chain for every AI use, with a named business owner and an exact human stop point. It must also match each use to technical bounds that can widen only when the evidence supports them. The Access Ladder shows how to set those bounds rung by rung, so the CEO and CIO can separate the business decision from the system access needed to carry it out.
How much access to your systems an AI agent needs, rung by rung. Free, with a PDF.
Questions CEOs ask about this
Which AI risks do I own as CEO, and which should I delegate to the CIO or CTO?
The CEO owns AI risk trade-offs within the board’s limits; the CIO owns technical boundaries. The CEO must balance unsafe use against delay, or risk unseen workarounds and lost business. Business leaders still own each use. The stake is clear ownership before AI acts across company systems.
Which risks are strategic rather than technical?
A risk is strategic when it changes what the firm will accept for a given return. That includes the harm from unsafe use and the cost of moving too slowly.
Founder, GRAIL
Johan has spent fifteen years in management consulting and executive leadership, including as Chief Product Officer of a SaaS company. He founded GRAIL to help European leadership teams work out what AI changes at the core of their business. He is the author of The Execution Revolution and the host of the ThinkRoom podcast.
Related reading
Sources
- The Judgment Walks Out: how a specialty insurer can speed up standard quotes while keeping human judgment on complex risks. (accessed October 8, 2026)
- The EU AI Act checklist: the board’s risk role, use-case ownership and the need for an exact human stop point. (accessed October 8, 2026)
- What does an AI-augmented IT department look like?: IT’s role in access, permissions and approval controls. (accessed October 8, 2026)
- AI agent access to your systems: why access should widen through clear steps, with proof before each wider action. (accessed October 8, 2026)
- What does an AI-augmented management team look like?: why key choices stay with the people who carry the result. (accessed October 8, 2026)
- What does an AI-augmented legal and compliance department look like?: Legal’s advice role and the business owner’s final call. (accessed October 8, 2026)
- The AI leadership gap is already ridiculous: how hard controls can push staff into tools the firm cannot see. (accessed October 8, 2026)
Next scheduled review: February 5, 2027