
Secure the AI tools and robots you use
Every AI system you deploy is also software that can be attacked or misused. The goal is simple: limit what it can reach, watch what it does, and keep a human in charge of anything that matters.
Know the main risks
| Risk | What it looks like | First defense |
|---|---|---|
| Prompt injection | Hidden instructions in a web page, email or file steer your AI agent | Limit tools and permissions; require approval for actions |
| Data leakage | Staff paste customer data or secrets into an unapproved chatbot | Approved-tools policy; data classification |
| Over-privileged agents | An agent can delete, pay or email without checks | Least privilege; human sign-off |
| Stolen API keys | Your AI account is used by criminals | Secrets manager; spend caps; alerts |
| Unsafe physical actions | A robot moves where people are | Safety assessment; emergency stop; zones |
Rules for AI agents
- Give each agent the smallest set of tools and data it needs.
- Separate "can read" from "can act." Reading your code is far less risky than running commands.
- Require a person to approve payments, deletions, external emails and production changes.
- Log every action with the input that triggered it.
- Treat content the agent reads from the internet as untrusted.
Rules for robots and connected machines
- Get the vendor's safety assessment and stop-distance data before deployment.
- Put robots on a separate network segment. Require signed software updates.
- Ask what cameras and microphones record, where it goes and how long it is kept.
- Make sure every worker knows how to stop the machine.
Use a framework: NIST's AI Risk Management Framework and the OWASP Top 10 for LLM applications are free, widely used starting points for governance and technical controls.
Sources and official resources
Updated November 17, 2026. This guide is general information, not legal or professional advice.