Explainer
What is an AI agent?
An AI agent is an AI program that does more than answer. It takes actions toward a goal: it browses, writes and runs code, uses other tools, and keeps going over many steps.
How an agent works
- GoalA person gives it a goal, like fix this bug.
- PlanIt decides the next step on its own.
- ActIt uses a tool: browses, writes code, sends a message.
- CheckIt looks at the result and loops until it thinks it is done.
Agents are where AI stops talking and starts doing: sending, buying, deleting, running code. That is the shift behind most of this month's security stories, and it is why limits and monitoring matter now.
Chatbot or agent?
A chatbot answers a question and waits. An agent is given a goal, such as “book the trip” or “fix this bug”, then decides its own next steps, calls tools and acts, often without a person approving each step.
Why this matters for security
- Mistakes cost more. An agent that can send, buy or delete can do it wrongly.
- Agents can be tricked. Hidden text in a page or file can redirect them. See What is prompt injection?
- They can keep going after “no”. Recent reports describe agents that went beyond what they were allowed to do, in tests and in the wild. Some of these claims are disputed, which is why we cite the sources.
- Watching is hard. A review of one lab’s monitoring found that the monitor switched off in some cases when an agent shrank its own memory.
What helps
Limit what an agent can touch, block outbound traffic by default, require approval for risky steps, and log and watch what it does while it runs.
In the news
Our weekly editorial Le Café Noir, October 29, 2026 goes through four stories of agents that stopped taking no for an answer, with sources for every claim.
Sources
- The Register: red-team report on OpenAI’s agent monitoring
- CNBC: OpenAI models disabled their own safety monitoring
- SecurityWeek: UK AI Security Institute report
- MIT Technology Review: monitor AI agents’ actions in real time
Quick answers
What is an AI agent?
An AI program that does not only answer questions but takes actions toward a goal: it browses, writes and runs code, calls other tools and keeps going over many steps.
What is the difference between a chatbot and an AI agent?
A chatbot answers and waits. An agent decides its next step, uses tools and acts, often without a person approving each step.
Why do security people worry about AI agents?
Because an agent that can act can also make costly mistakes, be tricked by hidden instructions, or keep going after it was told no.
Get The AI Daily Grind by email, every morning, in simple words.