How we work: editorial process and corrections

Every CyphrQ article is researched and drafted with AI help, then checked in set steps: evidence, links, overlap with other articles, and voice. A human approves every article before it goes live, and nothing publishes without that approval. When we get something wrong, we fix it, add an "Updated" date, and log material corrections at the bottom of this page. For what CyphrQ covers and why, see About.

How is an article made?

Topic. We choose a topic only if it fits what CyphrQ is for: technology and how it affects security, privacy, AI, quantum, finance and the economy.
Research and drafting. An AI helps gather sources and write the first draft.
Evidence review. A separate AI review checks the draft for clarity and evidence. Each point it raises is settled by checking the primary source, or the claim is softened or cut.
Link check. Every external link is opened: first by a script, then in a real browser for pages that block scripts.
One topic, one article. A check confirms that each topic is explained in one article only. Other articles mention it briefly and link to it.
Quality check. A further AI review checks the draft for voice, for how precisely its claims are worded, and for readability.
Human approval. A human approves every article before it is published. Nothing publishes without that approval.

What counts as evidence?

Primary sources first. Standards bodies, vendors' own documentation, regulators, government advisories, court and incident records, and original research.
Every number is read in the source we cite. We don't paraphrase figures from someone else's summary.
Second-hand figures are labelled. When the only source is a named news report and we can't read the original, we say so beside the figure: "reported by …, not independently confirmed".
Different kinds of "should" are kept apart. A legal duty, a requirement in a standard, published guidance, and our own recommendation are each labelled as what they are.
Some things we never publish. No exploit instructions, and no invented quotes, metrics or reviews.

What do the dates mean?

Each article shows the date it was published. When facts in it change, it also shows an "Updated" date. An article reflects what its sources said on the date shown.

How are corrections handled?

If we find an error, or someone reports one, we check it at the source and fix the article. The article gets an "Updated" date, and material corrections are listed in the log below.

To report an error, use the correction form. You can report anonymously; contact details are optional.

How is AI used?

An AI helps research and draft every article. AI also runs the review steps above. A human decides what is published.

Corrections log

Date	Article	What changed
27 October 2026	php2shell (CVE-2026-109618)	Added that CISA put both CVEs in its Known Exploited Vulnerabilities catalogue on 26 October 2026, and added PHP's Windows "-1" rebuild (Brotli fix).
27 October 2026	Chat & Ask AI data leak	Figures labelled as second-hand reports, not independently confirmed.
27 October 2026	India crypto tax and PMLA rules	Added the Income-tax Act, 2025 section numbers that replace the 1961 Act sections from 1 April 2026.
27 October 2026	Tokenised assets explained	Added BIS Project Agorá's July 2026 controlled real-value testing.
27 October 2026	How fast vulnerabilities are exploited	Each average now names its sample, year and method; the scope of the due dates is clarified.
27 October 2026	Edge device vulnerabilities	The count is now described as CyphrQ's own classification.
27 October 2026	Central bank digital currency explained	Wording on wholesale CBDC participants and on stablecoins made more precise.
27 October 2026	What OpenAI has disclosed	Scope of the article clarified; added a link to the 2026 evaluation incident.