What to do next
Show the working, step by step
Where I used this
Ask Shafaat, the AI assistant on this site, tells you it is an AI, answers only from my own writing, keeps an anonymous log of every question and answer, and has an off switch that hands visitors to the guided advisor. That is the minimum I would ask of any AI that talks to customers.
See how Ask Shafaat handles questions →Everything is calculated in your browser. Nothing you type is sent or stored.
How the four tiers work
- Prohibited. Manipulation that causes harm, exploiting age, disability or money worries, social scoring, and reading emotions at work or in school. Banned in the EU since 2 February 2025.
- High-risk. Hiring and work, education, credit scoring, life and health insurance pricing, access to public benefits including health care, emergency triage, critical infrastructure, policing, migration, courts and elections, some biometrics, and AI inside regulated products such as medical devices.
- Transparency. Chatbots must say they are AI, and AI-generated content must be marked. In force since 2 August 2026.
- Minimal. Most internal and marketing uses. No AI Act duties, though the operating rules in your result still pay for themselves.
Build once, for the strictest market you serve
The EU rules follow the user, not the headquarters: a team in Dhaka or Detroit serving people in the EU is in scope. Colorado’s rewritten law adds notice, an explanation, data correction and human review for consequential decisions from 1 January 2027. The cheapest path is one design that passes both: a named human owner, a log of every decision, a way for people to appeal, and an off switch. Treat the deadline as a design brief, the argument of my essay on the AI Act.
Get new free tools by email. One short note when a new tool or essay goes live. No spam, unsubscribe any time.
Shipping AI into a regulated workflow?
Bring the use case to a free 30-minute call, and I will help you design the oversight in before launch rather than bolt it on after.
Questions
Is my AI system high-risk under the EU AI Act?
It is high-risk if it is used in an Annex III area, such as hiring, education, credit scoring, life and health insurance pricing, access to public benefits including health care, emergency triage, critical infrastructure, policing, migration, courts and elections, or some biometrics, or if it is a safety part of a regulated product such as a medical device. A narrow helper task that does not profile people can be exempt, but the provider must document why.
When do the EU AI Act high-risk rules apply?
After the Digital Omnibus, in force since 27 July 2026, rules for standalone high-risk systems apply from 2 December 2027 and for AI inside regulated products from 2 August 2028. Bans have applied since 2 February 2025 and transparency rules since 2 August 2026, with until 2 December 2026 to mark AI content from systems already on the market.
Does the EU AI Act apply to companies outside the EU?
Yes. It applies when a system is placed on the EU market or its output is used in the EU, wherever the company is based. A team in Bangladesh or the US serving people in the EU is in scope.