74% think they'd pass an AI audit. *CIOs are liable* for systems they never approved.

The confidence gap

Kovrr's research found that 74% of organisations believe they could pass an AI compliance audit today ^[1]. At the same time, CIOs report being held accountable for AI systems they never knew existed, let alone approved.

It's like thinking you've locked all the doors when half the windows are open and you didn't know the building had a second floor.

The gap isn't a failure of intention. It's what happens when AI tools are easy to adopt and hard to track. A marketing team runs a chatbot trial. A recruiter tries a CV screening tool. An underwriting department tests a pricing model. Each decision feels local and low-risk, until someone asks the CIO to show what the organisation's AI systems have been doing for the past six months.

Where the systems actually are

The CIO answers for systems IT never procured, because the AI Act and other regulations don't care which department pressed the button. Under Article 26, the organisation using a high-risk AI system has duties: use it according to instructions, keep humans in the loop, monitor its operation and keep the logs ^[3].

Those duties apply whether the system was bought centrally or subscribed to on a corporate card. For high-risk systems under Annex III, they apply from 2 December 2027 ^[2].

The first question isn't whether you're compliant. It's whether you know which AI systems are running.

What the law asks for

Article 12 of the AI Act says high-risk AI systems must technically allow for automatic recording of events over their lifetime ^[4]. The logs are for traceability, for following up after market release and for the user's own monitoring.

Article 19 says providers must keep logs they control for at least six months. Article 26.6 says the same for users ^[5]. The law doesn't say how to protect the logs, just that they must exist and be kept.

That still leaves a question: if a supervisor or an auditor asks to see what a system did six months ago, and the log is kept by the same people who could edit it, how do you show it's the record from the day?

The part nobody talks about

A log file is like a diary. It says what happened, as long as you trust the person holding the pen.

Governance tools can tell you which systems are running and which logs exist. That's the first step, and it matters. The second step is harder: making those logs hold up when someone questions them.

Access controls and monitoring are the foundation, but they don't solve the last-mile problem. When the people who keep the logs could also change them, how do you show that nobody did?

How a digital witness works here

The thing to seal is each log record as it's written, or log batches every few minutes: the event type, the timestamp, the system version and the user or session involved. Sealing takes a fingerprint of the record, not the record itself, so nothing leaves your systems.

You seal through WitniumChain, using the REST API, the SDK or the MCP server. When a log is questioned, anyone you share it with can calculate the fingerprint again and check it against the sealed one. If it matches, the log hasn't changed since it was sealed. Change a single character and it won't match.

Auditors, supervisors and your own incident responders can check the logs. The witness shows that a record is exactly what was written at that time. It doesn't show that the log is complete, that the decision was right or that the system was lawful.

The first 2,000 witnesses a month are free. Read the documentation to get started.

What this means for you

If you're responsible for AI governance, here are four questions to ask this week:

1. Do we have a list of AI systems currently in use, including those procured outside IT?
2. For high-risk systems, who is responsible for the logs and where are they kept?
3. Could we show an auditor what a system did six months ago, in a way the auditor could independently check?
4. If a business unit wants to trial a new AI tool, does the approval process include the CIO or whoever will answer for it later?

The gap Kovrr found isn't about bad governance. It's about governance that hasn't caught up with how easily AI gets deployed. Closing it starts with knowing what's running, and making sure the logs can speak for themselves when the questions come.

What to watch

2 December 2027: High-risk requirements under Annex III of the AI Act apply, including the logging duty in Article 12 and the user duties in Article 26 ^[2][3].

2 August 2028: High-risk requirements under Annex I apply ^[2].