Large Action Models

What are Large Action Models?

Large Action Models (LAMs) are AI systems designed to turn natural-language goals into sequences of actions—clicks, API calls, or device controls—beyond text generation alone.

Action-taking systems inherit agent risks: excessive agency, prompt injection via UI/tool content, and irreversible side effects. Red-team tool permissions before wide rollout.

Related Giskard articles

• OWASP Top 10 for agentic applications 2026
• Function calling: testing agent tool usage
• Excessive agency failure case

Red-team action-taking agents

Probe LAMs for unsafe tool use and goal hijack with Giskard continuous red teaming. Learn more.

Further reading

Authoritative reference: OWASP Top 10 for Agentic Applications.