Home / Technology / FBI Veteran Warns: Your AI Is Out of Control

FBI Veteran Warns: Your AI Is Out of Control

Summary

  • Many companies lack baselines for normal AI agent activity.
  • Rogue agents can cause significant damage before detection.
  • Prioritize trained staff over more cybersecurity tools.

Many enterprise security teams are ill-equipped to identify when an AI agent deviates from its intended function. This deficiency stems from a widespread failure to establish baselines for normal agent behavior within their networks. Paul Bingham, a former FBI special agent and now an executive dean at Western Governors University (WGU), highlights this critical gap.

Bingham likens granting AI agents broad access without proper restrictions to giving a child the keys to a car. He advises that businesses must implement three key measures: restrict agent access, define and monitor normal agent activity for deviations, and prioritize investing in trained security personnel over additional technology tools.

The threat landscape is evolving, with cybercriminals adopting AI tools previously used by nation-state actors. High-profile breaches, such as the one at Hugging Face, underscore the emerging risks posed by AI agents penetrating production infrastructure. The distinction between criminal and state-sponsored attacks is blurring, emphasizing the need for robust defenses against both.

Detecting rogue AI agents necessitates understanding typical activity patterns. While many companies claim to have baselines for general network activity, this is less common for AI agents. Bingham suggests that most companies may be overestimating their ability to recognize anomalous agent behavior. He also advocates for integrating security professionals into the AI development process from the outset.

WGU is addressing the demand for AI expertise by launching new bachelor's degree programs in AI engineering, infrastructure and operations, and security. These programs aim to equip graduates with the skills needed to manage and secure AI systems effectively, preparing them for evolving roles in the cybersecurity field.

Disclaimer: This story has been auto-aggregated and auto-summarised by a computer program. This story has not been edited or created by the Feedzop team.

Read more news on

Property Code: 5571