How we use AI

Where AI helps, where it does not, and the checks around it.

Right now: AI summaries are switched on.

What AI does

• Summaries: an AI model reads the headlines and excerpts from every source that covered a story and writes a short summary at three levels: Simple, Standard and Expert.
• Ask: when Ask is switched on, an AI model answers questions about our stories, using only stories from our own site as its material.
• What to do: for stories about vulnerabilities, AI may suggest steps, which are checked against the vulnerability data before they are shown.

What AI does not do

• It does not choose which stories we publish or how serious they are. Severity comes from CVSS scores and CISA's list of exploited vulnerabilities.
• It does not write headlines. Headlines come from the sources.
• It cannot browse the web, run tools or take any action. It only receives text and returns text.

The model

We use Claude models from Anthropic, called directly from our own servers. Publisher text is sent to the model only to write the summary and is deleted when the job ends. We never send readers' personal details to the model.

Checks

• Every summary says it was written by AI and links to the sources it was written from.
• Summaries are checked automatically for links, markup and text that looks like an attempt to manipulate the model. Anything suspicious is held for an editor.
• Editors can rewrite or pull any summary, and readers can report one with "Wrong?".

When AI is off

If AI is switched off, stories still appear, with the source headlines and links but no summary.

Last updated 2 November 2026