AI Security

AI Security
Security of artificial intelligence systems and their dependencies

AI security encompasses protecting AI systems, data, models, infrastructure, and interfaces against malicious or unauthorized activity. In physical settings, cyber incidents can also affect sensors, actuators, operational decisions and potentially physical processes. Security overlaps with, but is distinct from, AI safety.

Attack surfaces

Relevant attack surfaces include training or update data, model-serving endpoints, connected devices, privileged APIs, software dependencies, and communication links. Unauthorized changes to configurations or sensor data may alter an AI system's behavior.

Security measures

Organizations use conventional controls such as access management, network segmentation, logging, patching and key management alongside AI-specific evaluations and monitoring. NIST's risk framework identifies security and resilience among the characteristics of trustworthy AI.

Physical systems

For robots and industrial automation, security plans must take account of real-time operation, recovery requirements and the consequences of interrupted commands. Operational-technology guidance offers additional context for industrial environments.

Further reading

• Guide to Operational Technology (OT) Security, SP 800-82 Revision 3 — NIST

See also

• AI Safety
• Robot Cybersecurity
• Zero Trust Security
• Digital Identity
• Physical AI Identity