Skip to content
Nubent

/IBM · Jul 30, 2025

IBM: AI breaches without access controls

IBM's Cost of a Data Breach 2025 report measured AI security for the first time: 13% of organizations reported breaches of AI models or applications, and nearly all of them lacked AI access controls.

What changes

Among organizations that suffered a breach of their AI models or applications, 97% said they had no AI access controls. One in five organizations also reported a breach caused by shadow AI, meaning AI tools people use without approval, and 8% don't know whether their AI systems were compromised. The study is based on 600 organizations, and IBM notes that those reporting AI breaches are still a small part of the sample.

How to implement it

  1. Map which AI tools people use, approved or not, and with what data.

  2. Define who and what can call each model or agent, using IAM roles instead of shared keys.

  3. Log every model call so you can investigate an incident.

  4. Write a simple AI usage policy and enforce it with controls, not just a document.

Nubent's take

The headline number is 97%, but read it carefully: it covers companies that already had an AI breach, not everyone. It still shows the most common pattern: AI gets adopted fast and access controls arrive late.

Next step

Should we implement it in your account?

We check whether it fits, how much work it takes and what risks it carries. 30 minutes with an engineer.