Intent-Based Security Tools Promise to Spot Bad Actors Early, But What Happens When the AI Gets It Wrong?

A new wave of security products is chasing an idea borrowed from the film Minority Report: stopping harm before it happens. The pitch is that a small AI model runs on every laptop and watches what people do across their apps, browsers and AI tools, much like a colleague looking over their shoulder. It tries to understand what each person is attempting, decide whether they are malicious, and step in before damage is done.

The same monitoring also keeps a complete record of work, so an analyst can read what happened instead of piecing it together from scattered logs. The scenarios named include an employee pasting sensitive data into an unapproved AI assistant, an outsider logged in with stolen credentials, or someone following a scammer's instructions to hand over remote control of their laptop.

Sophos notes that the watching part is now straightforward with local models. The harder question is design: what should a system do with a judgement about intent that will not always be correct? The article draws on the film's lesson. The Precrime system worked only while its three precognitives agreed, and it failed when it deleted the dissenting "minority report" and removed its own doubt. This is only the opening of the original piece, so the practical answers are not covered here.