Security startups race to rein in AI agents that act on their own

A new wave of security companies is trying to keep AI agents in check, according to The Information. AI agents are programs that take actions for you, like sending emails or moving files. The companies say they can watch what these agents do and set limits on it.

What we know

The Information describes a growing group of security firms built around this one problem. The idea is simple. If a program can act on its own, someone should be able to see its actions and stop the ones that go too far. The full article was not available to us, so the names of the companies and how their tools work are not covered here.

Also in the report

The same column carries a separate heads up. New OpenAI models could arrive as soon as Friday, according to people with knowledge of them. They could include the next generation of its Astra, Sol and cyber-focused models. The report notes that model timelines can change.

What is not known yet

• Which security companies are involved
• How well their tools work in real use
• What they cost
• Whether the OpenAI models will launch on that date

What to do

If your business lets an AI tool act for you, ask two questions. What can it access, such as email, files or accounts? And who is watching what it does? Start with limited access and widen it only when you can see its actions.