EU AI Act
LetterForge and the EU AI Act
What LetterForge does, what it doesn’t do, and where we think it stands under the EU’s AI Act.
What LetterForge does
- It reads the email thread with a candidate from your mail app, on your computer.
- It drafts emails to candidates and clients: when you click, and on Solo and Team also by itself as soon as a candidate replies.
- The draft opens in your mail app. You read it, change it and send it yourself.
- It keeps a board per vacancy where you move candidates along yourself, and reminds you to follow up.
What LetterForge never does
LetterForge helps you write. Everything else stays with you.
- Score, sort or filter out candidates
- Search the internet for people
- Build personality profiles
- Record conversations
- Send anything without your click
Candidates and emails stay on your computer. For a draft, only the part of the thread it needs goes through our server to OpenAI (USA), with signatures, phone numbers and addresses taken out.
Why we don’t consider LetterForge a high-risk system
Annex III, point 4(a) of the AI Act lists as high-risk the AI systems intended to be used for the recruitment or selection of people, in particular to place targeted job advertisements, to analyse and filter job applications, and to evaluate candidates.
LetterForge does none of this. It places no ads, doesn’t analyse or filter applications and evaluates no one. You decide whom to approach, present to a client or turn down, and LetterForge then writes the email for it.
Under Article 6(3), an Annex III system that profiles people is always high-risk. LetterForge assesses no personal traits and builds no personality profiles.
For these reasons we don’t consider LetterForge a high-risk system under Annex III. This assessment covers the features as of the date above.
Transparency: AI drafts, a person sends
- Every email LetterForge drafts is an AI draft. You see all of it before anything goes out, and you send it yourself from your mail app, from your own address.
- LetterForge doesn’t talk to candidates itself. There is no chat for applicants and no automatic sending.
- What goes out for a draft and what stays on your computer is set out on the security page.
Deadlines of the AI Act
The dates of the AI Act as amended by the Digital Omnibus on AI, checked against the Official Journal of the EU.
-
The prohibitions in Article 5 apply, among them emotion recognition in the workplace. So does Article 4 on AI literacy, which the Digital Omnibus has since reworded.
Art. 113, third paragraph, point (a) AI Act; Art. 4 replaced by Art. 1(5) of Regulation (EU) 2026/1744
-
The Digital Omnibus on AI enters into force. It postpones the obligations for high-risk systems.
Art. 4 of Regulation (EU) 2026/1744, published on 24 July 2026
-
The transparency obligations of Article 50 apply. Systems already on the market before then have until 2 December 2026 for the machine-readable marking under Article 50(2).
Art. 113 AI Act; Art. 111(4), inserted by Art. 1(39) of Regulation (EU) 2026/1744
-
The obligations for high-risk systems under Annex III apply, including point 4 (employment and workers’ management). The original date was 2 August 2026.
Art. 113, third paragraph, point (c) AI Act, as amended by Art. 1(40) of Regulation (EU) 2026/1744
Sources
We are happy to answer questions from you, your client or your IT.