An AI agent is a software system that uses AI to pursue a goal and can take actions through tools, such as searching files, using a browser or running approved commands.
How it works
An ordinary chatbot may only answer a question. An agent can be given a task and use connected tools to carry out steps. The actions it can take depend on its configuration, access and approval controls. Agents do not all operate without human review.
Why it matters
An agent can mishandle sensitive data, follow malicious instructions found in material it reads, or use permissions more broadly than intended. Those risks depend on what the particular agent can access; using an agent does not itself mean an account has been compromised.
What you can do
Give agents only the access needed for the task, review important actions and treat instructions from untrusted files or websites as data rather than authority. Monitor connected accounts and remove access that is no longer needed.
For more detail, NIST describes AI agent systems and their ability to act. Read how stolen AI keys can support cyberattacks for a related Hacked.com report.
