Security for AI
Identity, data, and abuse cases for models and agents, plus detection that treats an agent as a principal. AI inside the security workflow stays on a leash too.
For Security leads whose estate now includes models, agents, and the data they can see.
Outcomes
- Every agent has an identity, a scope, and an expiry
- Prompt, data, and tool abuse cases tested
- Detection when an agent acts outside its line
What is included
- Threat model for the AI workflow
- Agent identity and access reviews
- Guardrails, logging, and secrets
- A tabletop: the agent does the wrong thing
Platforms: IAM, SIEM, Cloud security, DLP
Start with this