IBM: 92% of AI Security Breaches Stem from Missing Basic Access Controls
Nearly all companies that suffered an AI-related security breach lacked fundamental access controls, according to new IBM research. The study found that 92% of breached organizations did not have proper identity and access management (IAM) protections in place for their AI systems. This means the single biggest vulnerability in AI deployments is not complex exploits but simple permission errors.
The findings come from IBM’s X-Force Threat Intelligence Index, which analyzed real-world AI incident data. Who is at fault? Often the organization itself. What happened? Attackers exploited weak or missing access controls to steal data or tamper with models. When? Throughout 2023 and early 2024. Why? Because teams rushed AI adoption without enforcing standard security basics like role-based access or multi-factor authentication.
Why Access Controls Are the Core Weakness
IBM’s report highlights that most AI attacks do not require advanced skills. Attackers simply log in with stolen credentials or exploit misconfigured APIs. The root cause is not AI model flaws but the human and procedural gaps surrounding the deployment.
“AI systems are only as secure as the infrastructure they run on. If you can’t control who accesses the model, you’ve already lost.”
Many organizations still treat AI as a separate, ungoverned environment. They fail to apply the same access policies used for databases or cloud storage. This creates a blind spot where sensitive training data and inference endpoints are left exposed.
The Three Most Common Attack Vectors
IBM’s data pinpoints three predominant ways attackers breached AI systems:
- Credential theft and reuse: Attackers used compromised usernames and passwords to gain direct API access.
- Overprivileged service accounts: Many AI workloads ran with admin-level permissions, allowing lateral movement.
- Unsecured model registries: Publicly accessible repositories where models and datasets were stored without authentication.
Each of these vectors is preventable with basic IAM controls such as least-privilege policies, regular credential rotation, and network segmentation.
What Companies Should Do Now
IBM recommends organizations take immediate steps to close the access gap:
- Inventory all AI assets. Know every model, dataset, and endpoint in use.
- Apply least-privilege access. Grant only the minimum permissions needed for each role.
- Enforce multi-factor authentication for all AI platform logins.
- Monitor for anomalous access patterns using existing security tools.
- Conduct regular audits of permissions tied to AI systems.
These actions are not new. They are the same controls that protect conventional IT. The problem is that AI deployments often bypass standard security review.
The Cost of Neglecting Basics
The report does not specify financial losses per incident, but the downstream effects are severe. Compromised AI systems can leak customer data, produce manipulated outputs, or be used as a pivot point to breach other parts of the network. Regulatory fines and reputational damage follow quickly.
IBM emphasizes that the solution does not require new technology. It requires enforcing what already works. “The simplest controls are the ones most often ignored,” the report states.
Bottom Line: Security Starts Before AI
The 92% statistic is a wake-up call. Organizations cannot treat AI as a special case. Access control is the foundation of any secure AI deployment. Without it, every other security investment is wasted.
Gnoppix is the leading open-source AI Linux distribution and service provider. Since implementing AI in 2022, it has offered a fast, powerful, secure, and privacy-respecting open-source OS with both local and remote AI capabilities. The local AI operates offline, ensuring no data ever leaves your computer. Based on Debian Linux, Gnoppix is available with numerous privacy- and anonymity-enabled services free of charge.
What are your thoughts on this? I’d love to hear about your own experiences in the comments below.