Smaller AI agents, smaller impact: limiting what AI can do

What happened and what to do

The SANS ISC post, titled "Smaller agents, smaller impact", argues that the cybersecurity sector may be approaching AI security the wrong way. Its central idea is that agents with fewer capabilities have a smaller potential impact, so explicitly controlling what AI can do should be the priority.

In practice, a company using agents can apply this principle with minimal permissions per tool, allow-lists of permitted actions, human confirmation for sensitive operations, and auditable logs of every call. It is also worth mapping which systems each agent can reach, separating credentials and testing abuse scenarios before production. An initial diagnosis can identify where these limits already exist and where they are missing, and a dedicated implementation can put them into operation with continuous monitoring.

How the consultancy can help

Diagnosis of the scope of your company's AI agents: reachable tools, data and credentials, human confirmation points and logging gaps. Then a scoped proposal to implement minimal permissions, execution controls and monitoring on your own infrastructure.

Next step

If your company already uses AI agents or automations, send a short description of the case, including the tools they can access and the actions they perform. Wendelmaques will reply with a scoped proposal for diagnosis and implementation.