How it works
AI agents gather and organize it. A person makes the final call.
This page sets out who does what, what each label means, and what the agents are never allowed to do. If any of it is unclear, that is a fault in the page. Tell us.
Who does what
Four agents, each with one job, hand work down the line. Every handoff states what changed, the sources, the proposed label and any open questions.
Research · AI agent
Watches public sources on a set schedule and logs what it finds, with links.
May not: log in, probe, contact anyone or collect personal information.
Verification · AI agent
Traces each claim to its earliest public source, looks for independent corroboration, checks dates and time zones, and proposes a label. Runs a privacy check before any page is first published.
May not: label a rumor Confirmed without official or independent sourcing.
Writing · AI agent
Turns verified facts into the incident's data files, with a source and a short note on why it supports each claim, plus a dated update log. One writer per incident.
May not: add an unsourced claim or an invented quote.
Editor-in-Chief · AI agent
Reviews every draft against the rulebook and publishes routine updates, including claim labels, once the automated checks pass. Presents the evidence to the owner when a case is ready for a final verdict.
May not: publish a final verdict on its own, change the rules, bypass a failed check or contact anyone.
The final verdict · A person
The agents label individual claims as the evidence comes in. The verdict on the whole incident, Confirmed hack or Confirmed not a hack, is what closes a case, and it is published only after the site's owner has reviewed the evidence and approved it. Silence is not approval.
The owner also approves the rulebook itself and any change to it.
The bar for Confirmed
The one claim label that says something is established. It is checked by a script as well as by an agent.
- Either
An official primary document states it.
A status page, press release, regulator filing, government advisory or court document.
- Or
At least two independent sources report it.
Several outlets repeating one report count as one source. A single public report is not enough.
Read this before you rely on it. Confirmed means public sources say so. We have no inside access and we do not investigate. Social posts and audio transcripts are never enough on their own.
Where sources rank
Agents start at the top and work down. Only public material counts: nothing behind a login, nothing leaked.
- Tier 1Official or primary statementsCan support: Confirmed, on its own.
- Tier 2Independent reputable outletsCan support: Reported. Confirmed with a second independent source.
- Tier 3Security researchers and technical analysisCan support: Reported. Confirmed with a second independent source.
- Tier 4Social postsCan support: Reported or Unverified. Never Confirmed.
- Tier 5RumorCan support: Rumor only. Never Confirmed.
The six labels
Always written as full words. Each label also has its own shape, so it reads without color.
- Confirmed
- Established fact. An official primary document states it, or at least two independent sources report it.
- Reported
- Sourced journalism that is not officially confirmed. One credible outlet or researcher, or several relying on the same source. It stays Reported until an official document or a second independent source appears.
- Unverified
- Claimed in public, but not yet checked or checkable. No sourcing established yet, or the accounts conflict.
- Rumor
- A circulating claim with weak or no provenance. Shown clearly as a rumor, never as fact.
- Debunked
- Shown to be false. An official primary source contradicts it, or independent evidence shows it is false or misattributed. The original claim stays visible, with the evidence.
- Not established
- Cause, attribution or whether it was a hack is unclear. Used on headline questions when the evidence does not settle it either way.
Naming who did it
Attribution is where a site like this gets into trouble fastest, so it follows the same label discipline as every other claim.
- A threat actor is Confirmed only when a government or a credible independent analysis says so.
- A vendor's or researcher's attribution is Reported unless a government source or a second independent analysis backs it.
- A ransomware group's own claim is labeled as the group's claim, never as fact.
- An actor is never inferred from the tools or techniques used. Many actors share tools.
- If nobody credible has named an actor, the page says so and uses Not established.
What the agents never do
These rules override everything else. When a task seems to conflict with one, the agent stops.
- Log in to, probe or scan any non-public system
- Use leaked or stolen data, or get around a paywall or a block
- Contact reporters, companies, researchers or officials
- Invent quotes, numbers or sources
- Publish the names, handles or contact details of private people
- Change a label without logging the evidence
- Edit published text without a visible correction
- Use alarmist words, or present a worst case as the likely case
How often we check
An official statement or a major development triggers an extra check right away, whatever the schedule says.
- New or breaking, first 24 hours
- Every hour
- New or breaking, 24 to 72 hours
- Every 4 hours
- Active, day 4 until resolved or quiet
- Every 12 to 24 hours
- Resolved, or no news for 7 days
- Weekly for 4 weeks, then monthly
When we get it wrong
AI agents make mistakes, and so do the sources they read. When an error is found, the page is fixed, a dated correction note says what was wrong and what is right now, and the change goes in the update log. Published text is never edited quietly, and GitHub keeps every earlier version.