The Guardrail covers the risk side of shipping AI: what regulation actually requires, what an auditor will accept as evidence, and what an attacker can do once a model holds credentials. Four beats, written for the people who have to sign something rather than the people selling the tooling.
The name describes the posture. We are not here to argue whether AI should be deployed. We are here to say, precisely, what has to be true before it is.
01Editorial standards
Compliance writing fails in a specific way: it summarises a summary of a law, and the summary was written before the law changed. We publish against the primary text.
- Every obligation names its provision. Instrument, article, and the date it applies from. If we cannot cite the provision, we do not state the obligation.
- Consultancy summaries are leads, not citations. We use them to find a development and then read the source ourselves.
- Dates carry a verification date. The compliance clock was last checked against primary sources on 2026-09-12.
- Corrections are visible. Substantive changes update the date on the briefing itself.
- Nothing here is legal advice. We report what instruments say. Applying them to your organisation is your counsel's job, not ours.
- Sponsored briefings are labelled at the top of the page and in the feed, and editorial staff have no role in producing them.
02How a claim gets verified
Before a briefing runs, every date, article number and framework reference in it is checked against the current text of the instrument. This is not a formality. While preparing this site we found three widely repeated deadlines that had already moved, including one that appears in a large share of published compliance calendars.
Where a claim rests on reporting we could not independently confirm, the briefing says so in its references section and names who published it first.
03Compliance clock
The clock runs down the right-hand side of the front page and every section page. This is the full list. Every entry was checked against the text of the instrument itself on 2026-09-12, not against a compliance calendar published by someone else.
| Obligation | Status |
|---|---|
| EU AI Act, Article 50 transparency duties | In force |
| EU AI Act, AI Office enforcement powers over GPAI | In force |
| EU AI Act, Article 50(2) compliance for systems placed on the market earlier | Dec 2, 2026 |
| Colorado AI Act, as revised by SB 26-189 | Jan 1, 2027 |
| EU AI Act, high-risk obligations under Article 6(2) and Annex III | Dec 2, 2027 |
| EU AI Act, high-risk obligations under Article 6(1) and Annex I | Aug 2, 2028 |
A date here changing is news in itself. When one moves, the briefings that relied on it get a correction note rather than a silent edit.
04Masthead
Three bylines write here, each with a defined beat. Every briefing names its author, and every author has a page listing what they cover and everything they have published.
Where a briefing carries an external reviewer, that person is named on the page. We leave the field empty rather than imply a review that did not happen.
05Advertise
The Guardrail carries display advertising and clearly labelled sponsored briefings. Available inventory: a leaderboard above the docket, an in-feed unit, and rail positions. There is no interstitial advertising, no autoplay media and no tracking wall.
Sponsored briefings are written to the same structural standard as editorial and are marked as sponsored on the page and in the feed. Ask for the media kit at editor@theguardrailreport.com.
Contact
Tips, documents, corrections and complaints go to editor@theguardrailreport.com. If you are reporting an error in a published briefing, quote the sentence and, where it concerns a legal obligation, the provision you believe we misread. That gets it fixed fastest.