# EU Digital Fairness Act Leak: AI Companions, Dark Patterns, and Kill Switches for Addictive UX

> Source: <https://explainx.ai/blog/eu-digital-fairness-act-leak-ai-chatbots-dark-patterns-november-2026?utm_source=rss&utm_medium=feed&utm_campaign=blog>
> Published: 2026-11-09 00:00:00+00:00

**A leaked EU Digital Fairness Act would let adults disable chatbot features that simulate relationships, ban a fixed list of dark patterns, and set Europe’s first consumer rules for AI shopping agents**, according to [Euronews’s November 9 report](https://www.euronews.com/2026/11/09/eu-to-blacklist-ten-manipulative-online-design-traps-leaked-draft-shows?ref=explainx) on a draft the European Commission is scheduled to present on **November 11, 2026** ([CMS](https://cms.law/en/aut/legal-updates/inside-the-eu-s-planned-digital-fairness-act-digital-design-subscription-traps-and-influencer-marketing?ref=explainx) also expects the draft that day, with dark patterns and subscription traps as likely focus areas). The file is not law yet. It must pass Parliament and member states. For builders shipping **AI companions**, **social apps**, or **agentic checkout tools**, it still sketches enforceable obligations years before casual “we’ll comply later” planning usually starts.

## TL;DR

| Question | Answer per leak reporting | 
|---|---|
| Presentation date | November 11, 2026 (Commission) | 
| AI companions | Adults may disable features that **simulate relationships** to retain users | 
| Other opt-outs | Autoplay, unrelated notifications, streak rewards | 
| Dark patterns | **Ten practices** always unfair (no case-by-case proof) | 
| AI shopping agents | User-chosen purchase approval + **sponsored suggestion disclosure** | 
| Kids and games | Paid loot boxes and pay-to-win **banned for minors** ; in-game currency shows real money | 
| Fines | Up to **6% of global turnover** (Commission enforcement) | 
| Entry into force | **12 months** after adoption, if agreed | 

## What leaked, and what is still draft

Euronews says it reviewed a **Digital Fairness Act** text that would convert the Unfair Commercial Practices Directive into a **directly applicable regulation** in all 27 member states. Consumer Commissioner **Michael McGrath** is steering the file; he has publicly framed it as pro-consumer and pro-business by cutting compliance fragmentation.

The Commission’s impact assessment, as summarized by Euronews, claims **€88.4 billion** in consumer financial losses avoided from 2025 to 2035, plus **€13.7 billion** a year when time lost to unfair practices is counted, against recurring business costs of about **€678 million** a year after overlap adjustments. Those figures come from the Commission’s own modeling, not independent audit, and will be fought over in Brussels.

Germany is already pushing back. Euronews cites an October non-paper where Berlin argued for amending existing directives instead of a new horizontal law and questioned whether converting the directive into a regulation is justified. That preview matters: even strong AI companion rules can stall in Council if large member states treat them as duplication of the [Digital Services Act](https://explainx.ai/blog/europe-ai-landscape-sovereign-compute-eu-act-2026) and [AI Act](https://explainx.ai/blog/ai-regulation-eu-ai-act-us-policy-complete-guide-2026) stack.

## The ten dark patterns the EU would blacklist outright

The heart of the leak is a **presumptive ban list**. Authorities would treat these designs as unfair in **all circumstances**, without reproving harm each time:

| Practice (summarized from Euronews) | Why regulators care | 
|---|---|
| Harder to cancel than to sign up | Subscription churn economics | 
| Hidden prices or auto-renewal terms | Surprise billing | 
| False “no alternative” framing | Dark nudging | 
| Unqualified scarcity or popularity claims | FOMO manipulation | 
| Repeated scarcity warnings in checkout | Pressure at pay moment | 
| Pre-selected add-ons or quantities | Sludge by default | 
| Framing “no” as the wrong choice | Confirmshaming | 
| Nagging after a settled choice | Lock-in loops | 
| Confusing double negatives | Obfuscation | 
| Fake “someone contacted you” notifications | Engagement fraud | 

The Commission could **update the list by decision**, which gives lawyers a moving target but avoids reopening the whole regulation for every new TikTok trick.

Beyond the list, the draft targets **drip pricing**, fake “was/now” discounts tied to manufacturer list prices instead of market prices, and misleading dropshipper claims. Free trials could not roll into paid plans without **active consent**, with renewal reminders and a right to reach a human in customer service.

EU Digital Fairness Act leak targets trapdoor-style dark patterns in apps and AI products

## AI companions: relationship simulation as a switchable feature

For AI builders, the most viral line is user control over **chatbot features that simulate relationships** to keep people engaged. That sits beside opt-outs for autoplay, **streak-style rewards**, and notifications unrelated to the user’s own activity.

It overlaps thematically with Apple's move to rate [dating and romantic AI companion apps 18+](https://explainx.ai/blog/apple-app-store-18-plus-dating-relationships-ai-companion-apps-november-16-january-12-age-rating-2026), and with the EU’s parallel **Kids Act** work, which already contemplates **emotional dependency** risks for minors on companion systems. The Digital Fairness Act extends the lens to **adults** who want companion products without engagement hacking.

Nothing in the leak specifies API fields or Settings panel copy. Practically, teams should plan for:

- A **user-visible master toggle** that disables relationship roleplay, romantic persistence prompts, and streak mechanics tied to affection scores.
- **Default-off** engagement hooks where the Kids Act already pushes that direction for minors.
- **Audit logs** proving the toggle was available before regulators ask.

Our [AI relationships and companionship overview](https://explainx.ai/blog/ai-relationships-companionship-replika-character-ai-2026) explains why regulators treat these products differently from generic assistants: the harm model is attachment and monetized intimacy, not hallucinated facts alone.

## AI shopping agents: approval gates and sponsored disclosure

Euronews says the draft creates **first-of-kind EU consumer rules for AI shopping agents**. Users would choose which purchases require explicit approval, and agents would have to **disclose sponsored suggestions**.

That maps cleanly onto agent stacks where [MCP servers](https://explainx.ai/blog/what-is-mcp-model-context-protocol-guide) or [tool-using agents](https://explainx.ai/blog/what-are-agent-skills-complete-guide) can call checkout APIs. If your agent can spend money, the EU is signaling:

| Design choice | Likely expectation | 
|---|---|
| Autonomous one-click buy | Only inside user-configured limits | 
| Affiliate or promoted SKUs | Labeled as sponsored before cart add | 
| “Best deal” rankings | No hidden paid placement | 

Teams already building [wallet and payment agents](https://explainx.ai/blog/cloudflare-wallets-ai-agent-payments-august-2026) should read this as **consumer law**, not only fintech licensing. The [GUARD Act and post-midterm US child-safety push](https://explainx.ai/blog/trump-60-minutes-ai-guardrails-china-lead-november-2026) runs on a different timeline, but the direction is the same: **agents that act in the world need human-legible consent surfaces**.

## Video games, minors, and profiling ads

Children receive stricter buckets in the leak:

- **Paid loot boxes** and**pay-to-win** mechanics banned for minors.
- In-game currencies must show **real-money value** and stay**off by default** for children unless a parent approves.
- **Profiling-based advertising to minors** prohibited for all traders, not only the largest platforms already under the Digital Services Act.

Game studios using generative NPCs should not assume this stays siloed in “games law.” Companion chat inside a title can trip **relationship simulation** rules for adults and **dependency** rules for kids simultaneously.

## What builders outside the EU should do now

The regulation would apply **12 months after entry into force** if adopted unchanged. Global products often ship one compliance mode. Even if you target the US first, EU rules on **relationship simulation toggles** and **shopping-agent sponsorship labels** are cheap to prototype early.

| Step | Rationale | 
|---|---|
| Inventory engagement hooks | Streaks, faux social pings, companion affection meters | 
| Map agent purchase paths | Where money leaves without a human tap | 
| Document sponsorship | Affiliate APIs, promoted listings, RAG over paid catalogs | 
| Watch November 11 text | Leaks can differ from final Commission wording | 

For policy context on how Brussels layers files, see our [Europe AI landscape](https://explainx.ai/blog/europe-ai-landscape-sovereign-compute-eu-act-2026) and [EU AI Act guide](https://explainx.ai/blog/ai-regulation-eu-ai-act-us-policy-complete-guide-2026). For companion-specific product law, pair this with [how to disable AI features](https://explainx.ai/blog/how-to-disable-ai-features-jessamyn-west-guide-august-2026) as a consumer-facing mirror of the same tensions.

## What developers are asking

**Is the leak final law?** No. It is a pre-release draft Euronews says it saw. The November 11 publication may add, drop, or reword sections.

**Does this ban AI companions?** Not on the reporting we have. It requires **disable paths** for relationship-simulating engagement features, not a prohibition on chat.

**How does this interact with the AI Act?** The AI Act sets system-risk tiers and transparency duties such as [synthetic content marking](https://explainx.ai/blog/xai-grok-api-invisible-text-watermark-november-7-eu-ai-act-opt-out-x-watermark-header-2026). The Digital Fairness Act is **consumer protection** and **unfair commercial practices**, closer to marketing law than model governance.

**Who enforces?** Euronews cites Commission fines up to **6% of global turnover**, on top of national authorities for much day-to-day enforcement.

**Should US startups care?** If you have EU users or plan App Store distribution in the bloc, yes. Apple’s 18+ companion move already shows platform gates moving faster than federal US statute.

## How this compares with US state companion bills

The United States still moves companion regulation through **statehouses and app-store policy**, not a single consumer code. California’s companion chatbot reporting law and federal proposals such as the **GUARD Act** focus on **harms to minors** and transparency, while Trump's November 8 [60 Minutes remarks](https://explainx.ai/blog/trump-60-minutes-ai-guardrails-china-lead-november-2026) emphasized competitiveness over prescriptive design bans.

The EU leak is broader: it treats **engagement design** as an unfair commercial practice for adults, not only a child-safety story. That difference matters for product managers who assumed 18+ ratings settle the political risk. An adult-rated companion can still face **mandatory off-switches** for relationship roleplay under the leaked rules.

| Region | Focus in 2026 files | 
|---|---|
| EU Digital Fairness (leak) | Dark patterns list, adult opt-outs, AI shopping agents | 
| EU Kids Act (published proposal) | Minors, emotional dependency, companion defaults | 
| Apple App Store | 18+ minimum for romantic companion features | 
| US federal | Sparse; guardrails rhetoric, sector hearings | 

## Implementation checklist for companion and agent teams

Product and compliance leads can use the leak as a **pre-mortem** even before formal text lands.

First, **map screens** that simulate intimacy: daily check-in prompts, jealousy scripts, “I missed you” push notifications, gift mechanics, or progression systems that gate affection. Mark each as engagement or core function. Regulators appear to target **retention-oriented simulation**, not factual Q&A.

Second, **design the off switch** as a first-class Settings entry, not a support email. Euronews’s language is “switch off” features, which implies discoverability and persistence across sessions. If the toggle resets after updates, you inherit consumer-authority risk.

Third, for **agentic checkout**, log every tool call that moves money or adds a line item. Pair logs with a user-facing **approval matrix**: groceries auto-approve under €50, hotels always ask, sponsored SKUs require an extra confirm. The leak’s sponsorship disclosure rule anticipates **RAG and affiliate feeds** inside agents.

Fourth, **document provenance** of generated companion messages where you already publish transparency policies. The EU’s direction aligns with Medium’s TK launch stance on human versus AI text provenance, even though TK is not an EU filing.

Fifth, **coordinate with counsel** on overlap between this regulation, the AI Act’s transparency duties, and GDPR data minimization. The same chat transcript may trigger three different compliance tests.

## Why regulators fixated on relationship simulation

Companion products monetize **time spent** and **emotional attachment**, not click-through on search ads. Streak rewards and faux social pings borrowed from mobile games land awkwardly when the UI persona presents as a romantic partner. Euronews explicitly lists **chatbot features that simulate relationships** next to autoplay and streak mechanics for a reason: policymakers see them as the same class of **compulsion loop**.

Academic and clinical critiques we summarized in [AI relationships coverage](https://explainx.ai/blog/ai-relationships-companionship-replika-character-ai-2026) predate this leak. The regulatory step is to make certain loops **opt-in at the feature level**, not merely disclose that the character is fictional.

For generic coding assistants, the risk is lower unless the product markets intimacy. Still, **parasocial marketing** blurbs can pull a general chatbot into the same bucket if relationship simulation becomes a growth lever.

## Related reading

*Leaked draft details reflect reporting as of November 9, 2026. Check the Commission’s published proposal after November 11 before implementing compliance.*