AI agent security basics

Threat classes

Prompt injection via documents and tickets that try to exfiltrate data or coerce tool calls.

Over-broad credentials—still the most common enterprise footgun.

Supply chain: models, plugins, and MCP servers you did not inventory.

Controls

Treat untrusted text as data, not instructions, when designing tool policies.

Least privilege, short-lived credentials, human gates on irreversible actions.

Monitor anomalous tool sequences the same way you monitor anomalous API use.

Document-borne attacks

Treat uploaded PDFs and ticket bodies as untrusted. Instructions inside documents should not expand tool permissions.

Separate "content to extract" from "instructions to follow" in the architecture, not only in the prompt text.

Incident response

Know how to revoke agent credentials, pause writes, and reprocess cases. Write the runbook before the first incident.

In practice

Map the workflow on a whiteboard before you open a framework: inputs, systems of record, humans, and irreversible writes. If that map is fuzzy, the agent will encode the fuzz.

Pick ten to fifty real historical cases as an eval set. Include the ugly ones. Run the agent offline against them until critical fields and hard rules are acceptable. Only then connect write tools.

Ship with a pause switch, a human queue, and a weekly review of override reasons. Promote repeated overrides into rules. That loop is how production systems improve—not another prompt brainstorm.

Common failure modes

• Treating a demo on clean samples as readiness for production volume.
• One shared service account with broad write access across systems.
• No owner for the exception queue, so failures pile up as noise.
• Changing prompts and models without regression gates on real cases.
• Measuring only model latency or thumbs-up, not completed-case cost and audit completeness.

What good looks like after ninety days

The first workflow is boring: stable override rate, known failure modes, operators who trust the queue. Config changes go through review. Traces answer "what happened to this case?" without archaeology.

At that point you can add a second document type or a second agent role. Expanding before the first path is boring is how programs stall with five half-built pilots.