PRE Security, which describes itself as a predictive artificial-intelligence security operations company, has released an updated version of its platform with new detection and response capabilities, and says it will show the product at Cyber Security World Asia 2024 at the Marina Bay Sands exhibition centre in Singapore. The company will exhibit with a local launch partner, NGEN Technologies.

The additions are a detection layer that watches endpoints and networks, and an assistant that lets an analyst interrogate data in ordinary language. The company says it has trained a model on security material so that it understands the context of an alert, and that features now sort and prioritise alerts automatically, reducing the manual work of investigating and writing up every one.

The gap between predicting and proving

The claim worth pressing on is the word predictive. A tool that flags a vulnerability before it is exploited is useful, but the release offers no measurement of how often it is right or how often it is wrong. Detection figures need a test set and a false-positive rate to mean anything, and neither appears.

The assistant raises a subtler problem. Automated triage speeds up the queue, but the analyst still has to check the summary before acting on it, and a language model that misreads an alert is harder to catch than one that simply misses it. The product may well work. The honest position is that a launch event is not a result: what a buyer should ask for is a trial against their own traffic, and a number for the alerts the system raises that turn out to be nothing. John Peterson, a co-founder of the company, said the aim is to respond to attacks before they reach the network. That is the promise; the test comes later.