August 2: The EU AI Act Grew Teeth
Chatbots must "disclose they are not human." Deepfakes must be labeled. AI-generated content must carry machine-readable watermarks.
This is not advice. It is a fine.
Three Things That Took Effect That Day
First, Article 50 transparency obligations
- AI that interacts directly with people (chatbots, voice assistants, AI agents, automated calls) must clearly tell users up front: you are talking to AI
- AI-generated images, audio, video, and text must carry machine-readable marks detectable as AI-made
- Deepfake content must be prominently labeled
The disclosure cannot hide in the footer or use vague wording. EU guidance explicitly pulls AI agents into scope — if you cannot predict whether an agent will interact with a human, disclose by default.
Second, GPAI enforcement powers
The European Commission can now investigate, fine, and order fixes for general-purpose AI model providers. The ceiling is 15 million euros or 3% of global annual turnover, whichever is higher.
Third, regulatory sandboxes
Member states must ensure at least one national AI regulatory sandbox runs, giving companies room to experiment.
A Common Miscalculation
Many Chinese firms assume "we operate domestically, so Brussels cannot touch us" — wrong.
If your AI product serves EU users, whether a website, an app, or an exported SaaS, it is in scope. Meta refused to sign the transparency code; X signed only the safety chapter — but fines do not care what you signed.
For systems already on the market, the machine-readable marking deadline was pushed to December 2. That is a grace period on one technical detail, not a waiver of the whole clause.
Do Not Misread the "Delay" on High-Risk Systems
The tough high-risk AI classification rules were postponed to December 2027 by the Digital Omnibus amendment passed in June. That is true.
But do not get comfortable — transparency (Article 50) and GPAI enforcement are hard rules effective August 2, with no grace period. Many firms misread "high-risk delayed" as "do nothing now." That is the trap.
If You Ship to Europe, Do These Five Things Now
- Add a prominent AI-identity notice in chatbot/agent UIs — no footers, no fine print
- Stamp AI-generated images, video, and text with machine-readable marks (C2PA-style standards are landing)
- Dual-label deepfakes (visible tag + machine-readable)
- Non-EU providers must appoint an EU-based authorized representative as the regulatory contact
- If you build frontier models, run systematic risk assessments and keep the records
Why This Matters to You
You might be building AI customer service, AI avatars, or AI copy tools and eyeing the European market.
This law is not here to scare you — it draws a "trust floor." Comply early and you build trust in Europe; ignore it and your product gets pulled, or you face a 3%-of-global-revenue fine.
Avatar-software teams should pay special attention: "the more human, the better" used to be the pitch; now "must say it is not human" is the floor. The entire product narrative has to flip.
Bottom Line
AI shifting from "human-like" to "must clarify it is not human" is a turn in the industry's underlying logic. Compliance is not a cost — it is a ticket to enter. Bake transparency into the product and you outpace rivals still pretending to be people.