Surface 10

AI / Agents

What lives here

Deployed models, agent permissions, prompts, training and inference data, integrations between LLMs and enterprise systems.

Primary threats
  • prompt injection
  • data leakage to model providers
  • over-permissioned agents acting beyond intent
  • model abuse / jailbreak
Defensive categories
  • AI security (model + agent + data + prompt governance)
  • DSPM for AI data
  • identity for agents and non-human identities