Uncovering AI Model Lineage for Security

New research reveals that country-of-origin labels can obscure AI model dependencies and inherited behaviors, potentially exposing security risks.

Bottom line: Security leaders must examine AI model lineage to ensure security and compliance.

What's happening: Researchers at Cisco have found that country-of-origin labels can obscure the upstream dependencies, inherited behaviors, and potential security risks of AI models. A recent study using the popular AI model, TensorFlow, found that 27% of AI models rely on components from China, with 15% of those components containing known vulnerabilities (CVE-2022-23841, CVSS score 9.3).

What to do: Security leaders should review AI model lineage to identify potential security risks and take steps to mitigate them, such as implementing robust security protocols and monitoring AI model behavior.

Source: SecurityWeek