AI security
The real risks of AI agents and tools: prompt injection, secrets, permissions, data protection and good practice.
5 articles
-
Run an AI agent in a sandbox
A sandbox limits the files, network connections, and resources an AI agent can access. Configure its isolation and test that restrictions work before use.
-
AI agents and the GDPR: what data leaves, and where does it go?
An AI agent can send your files, messages, and logs to several providers. Learn how to map these data flows and check the safeguards that protect your information.
-
Malicious MCP servers and skills: check before installing
Check the source, permissions, and behavior of an MCP server or skill before installing it, then test in an isolated environment without exposing your data.
-
Prompt injection: how an AI agent gets hijacked
Prompt injection turns external content into instructions for an AI agent. Learn how it works, what puts data at risk, and which defenses can limit the damage.
-
AI agent security: 10 risks and how to protect against them
AI agents can expose data, take unwanted actions, or exceed their budgets. Learn about ten concrete security risks and the safeguards that help reduce them.