Enterprise LLM Threat Atlas Highlights Key Risks and Defenses

Summary

The Enterprise LLM Threat Atlas identifies tool and agent abuse and supply-chain attacks as leading enterprise LLM risks, alongside prompt injection, over-permissioned connectors, knowledge-base poisoning and data leakage.

Key points

• The report covers self-hosted LLM applications and LLM features embedded in SaaS products.
• It identifies indirect prompt injection through retrieved enterprise content as a prominent threat.
• Over-permissioned connectors and copilots may enable sensitive-data discovery after a compromise.
• Tool and agent abuse and supply-chain attacks are highlighted as the most important current risks, based on the number of public reports and research cases found.
• The report describes a MITRE ATLAS and MITRE ATT&CK mapping approach and a prevent, detect and respond control baseline.