I know you don’t want them to want AI, but…
🔗 **Source:** [I know you don’t want them to want AI, but…](https://www.anildash.com//2025/11/14/wanting-not-to-want-ai/)
> [!note]+ Document Notes
> tags: [AI, Firefox, Mozilla]
> author: [Anil Dash]
> source: [AnilDash.com]
> reference: [https://www.anildash.com//2025/11/14/wanting-not-to-want-ai/]
> date created:
>
> # 🏷️ Title: I know you don’t want them to want AI, but…
>
<!-- deep-read:sections start -->
## 🧭 Table of Contents
- [📝 Summary](#-summary)
- [📖 Main Content](#-main-content)
- [🗒️ Notes](#️-notes)
- [📚 References](#-references)
- [🗣️ Paraphrase](#️-paraphrase)
- [📰 Newsletter](#-newsletter)
- [✅ Action Items](#-action-items)
- [🧪 Hypotheses](#-hypotheses)
- [📊 Methodology](#-methodology)
- [❓ Questions](#-questions)
- [🔮 Future Research](#-future-research)
- [💡 Implications](#-implications)
- [🧠 Learning Styles Assessment](#-learning-styles-assessment)
- [📊 Accessibility Assessment](#-accessibility-assessment)
- [🌍 DEIB Assessment](#-deib-assessment)
- [✅ Logical Validity](#-logical-validity)
- [🛡️ DARVO Analysis](#-darvo-analysis)
- [🎯 Conclusion](#-conclusion)
### 🔗 References
- Include links cited in or relevant to the article.
## 📝 Summary
- What happened: Anil Dash argues that while tech communities loudly reject AI in Firefox, hundreds of millions use AI daily; Mozilla should protect users and define a “good” AI path.
- Why it matters: Security, privacy, and user agency will be shaped by browsers; abstention cedes ground to worse actors.
- Key stakeholders: Mozilla/Firefox teams, open web/community, mainstream users, AI platform vendors.
- Risks and tradeoffs: Alienating anti-AI users vs. leaving AI users unprotected; scope creep vs. meaningful safeguards.
- Bottom line: Build privacy-first, user-controlled AI integrations, plus an “off” switch and plural Firefox options.
## 📖 Main Content
- Core arguments:
- AI use is mainstream; denial won’t protect users.
- Browser-level protections (like the pop-up wars) can mitigate harms when usage is inevitable.
- Scolding users won’t work; ship better, safer alternatives.
- Mozilla’s framing caused avoidable backlash; focus on vision, controls, and community.
- Context/definitions:
- “Big AI” harms: labor, consentless data use, environmental costs, erosion of trust.
- Analogy: early-2000s pop-up crisis solved by better browsers, not user shaming.
- Uncertainties:
- What exact features constitute “good AI” in-browser.
- Governance and defaults across different Firefox flavors.
### 🗒️ Notes
- “Hundreds of millions of users are choosing to go to these websites … and engage with these tools.”
- Regular people find AI media moving; tech elites’ tastes aren’t universal.
- Anti-AI forums overgeneralize; mainstream behavior differs.
- War on pop-ups analogy: browsers won by protecting users with better defaults (tabs, blockers).
- Ask: If people use ChatGPT, which browser best protects them?
- Strategy: Define “good AI,” build it, market it, and add a master “shut off AI” toggle.
- Positioning: “The best AI browser for people who hate Big AI.”
- Plurality: “There isn’t a Firefox—everyone is Firefox” (official, Zen, extensions, local LLMs).
## 📚 References
- How to protect your privacy from ChatGPT and other AI chatbots (Mozilla Foundation): https://www.mozillafoundation.org/en/privacynotincluded/articles/how-to-protect-your-privacy-from-chatgpt-and-other-ai-chatbots
- “I think nobody wants AI in Firefox, Mozilla” (Rodrigo Ghedin) [context to the debate]
- Mozilla Festival (MozFest): https://www.mozillafestival.org/
## 🗣️ Paraphrase
- A: Lots of people use AI by choice; ignoring that reality leaves them exposed to bad actors.
- B: Like pop-up blocking, browsers should lead with protections, plus a clear AI-off switch and multiple Firefox options.
## 📰 Newsletter
- This week: A pragmatic path for AI in browsers. Anil Dash says users already choose AI, so Firefox should protect them—privacy-first integrations, a global “AI off” toggle, and support for local models. Explore practical safeguards via Mozilla’s privacy guide and bring these ideas to your teams and meetups.
## ✅ Action Items
#### DevRel
- Publish demos of local LLM-in-extension workflows with strict data boundaries.
- Author guides: “AI Off by Default” patterns; telemetry minimization.
- Collect developer feedback on safe AI APIs (rate limits, red-teaming hooks).
#### OSPO
- Draft policy for AI feature contributions, decision rights, and sunset clauses.
- Add AI-related license/compliance checks for models/datasets.
- Set security intake for AI prompts/outputs, signing/attestation for models.
#### Community
- Host roadmap clinics on “good AI” features; run consent-aware prompt libraries.
- Pair contributors on privacy-preserving AI extensions.
#### WordPress
- Core AI: evaluation sets, safety rails, local-first pilots.
- Docs/Support: “AI Off” guides, provenance notes, content policies.
- Accessibility: review AI-generated UI text for WCAG risks.
#### The WPOCC
- Fund maintainers for AI-safety libraries; microgrants for local-first plugins.
### FAIR
- Enforce provenance checks, signatures for models; transparent funding and anti-capture covenants.
## 🧪 Hypotheses
- Browser-level AI privacy controls increase mainstream adoption of safer AI paths.
- A global “AI off” toggle reduces churn among anti-AI users.
- Marketing “best AI browser for people who hate Big AI” increases Firefox share among AI users.
- Local-LLM extensions reduce data exfiltration incidents vs. cloud defaults.
### 📊 Methodology
- Run A/B tests on onboarding with AI-off defaults; measure retention and trust.
- Compare telemetry of privacy incidents across local vs. cloud AI features.
- Survey panels on browser choice after safety messaging.
### ❓ Questions
- Which AI protections belong at the browser layer vs. extension?
- What default model governance is acceptable to the community?
- How to verify provenance and consent of training data?
### 🔮 Future Research
- Standard APIs for local model sandboxes.
- Usability studies for master AI toggles and per-site AI permissions.
### 💡 Implications
- Practice: ship opt-out/opt-in clarity, per-origin AI permissions.
- Policy: transparent datasets, red-teaming disclosures.
- Funding: support community-developed safety tooling.
### 🧠 Learning Styles Assessment
- Reading level: ~Grade 9–10 (Flesch-Kincaid estimate).
#### 👀 Visual
- Improve with architecture diagrams of browser AI data flows.
#### 🔊 Aural
- Add a 3-minute audio summary.
#### 🗣️ Verbal
- Glossary for “local LLM,” “provenance,” “attestation.”
#### 🧩 Kinesthetic
- Step-by-step: install a local-LLM Firefox extension.
#### 🧠 Logical
- Include decision tree: when to enable AI features.
#### 🧑🤝🧑 Social
- Community Q&A with moderation.
#### 🧘♀️ Solitary
- Reflection prompts: personal AI risk tolerance.
#### 🌿 Naturalistic
- Note environmental impacts of model choice, efficiency tips.
### 📊 Accessibility Assessment
- Screen reader: ensure headings/landmarks; add descriptive alt text.
- Keyboard: verify focus order/skip links.
- Contrast: maintain ≥4.5:1; check line-height.
- Reflow: support 200% zoom, reduced motion.
- Media: captions/transcripts; avoid autoplay.
- Forms: labels, inline validation, generous targets.
- Map to WCAG 2.2 AA.
### 🌍 DEIB Assessment
- Use global examples; link translations.
- Plain-language summaries; avoid idioms.
- Consistent structure and icon+text labels.
- Balanced representation in examples and imagery.
- Clear community guidelines and escalation.
### ✅ Logical Validity
- Strengths: Inductive analogy (pop-ups → AI harms) motivates browser-layer solutions.
- Possible fallacy risk: Hasty generalization about “hundreds of millions” without specific stats; fix by citing usage data.
- Tone: Pragmatic, gently contrarian, community-forward.
### 🛡️ DARVO Analysis
- Deny: Does not deny harms; acknowledges them explicitly.
- Attack: Critiques tactics (scolding), not people.
- Reverse victim/offender: Avoided; centers user protection and agency.
## 🎯 Conclusion
- Bottom line: People already use AI; build protective, user-controlled AI experiences in Firefox, with a universal off switch and plural options.
- Next steps: Define “good AI” criteria, implement privacy-first controls, market to safety-conscious users, and support local LLM paths.
## 🧱 Refactor Appendix (append after 🎯 Conclusion)
### 📚 Literature Split Pad
- Purpose: Stage Konik CEE units that become per-source Literature notes.
- Split level: H3.
- Include verbatim quotes under 25 words and a [view highlight] link.
### [H1] Browser protections make AI safer
> [!quote] [I know you don’t want them to want AI, but…](https://www.anildash.com//2025/11/14/wanting-not-to-want-ai/) by [[Anil Dash]] via [[AnilDash.com]]. Published 2025-11-14. Accessed .
> “Using AI tools is an incredibly mainstream experience now.”
Two lines: Use reality to justify protections at the browser layer.
[view highlight](https://www.anildash.com//2025/11/14/wanting-not-to-want-ai/#H1) #xref #articleseed ID:rw_[MISSING]
### [H2] Offer a master AI-off switch
> [!quote] [I know you don’t want them to want AI, but…](https://www.anildash.com//2025/11/14/wanting-not-to-want-ai/)
> “Just give people the ‘shut off all AI features’ button.”
Two lines: Increase trust and reduce backlash.
[view highlight](https://www.anildash.com//2025/11/14/wanting-not-to-want-ai/#H2) #xref ID:rw_[MISSING]
---
### 🔖 Atomic Split Pad (Zettel Candidates)
#### [Z1] Browser-layer AI controls reduce harm
- Applies when users engage AI via the web.
- Pointer: Dash article.
- Links: [[Browser-Safety]] [[Responsible-AI]]
#### [Z2] A global AI-off toggle increases trust
- Scope: Privacy-conscious user segments.
- Pointer: Dash article.
- Links: [[User-Agency]] [[Preference-Architecture]]
---
### ✏️ Fleeting Queue (Extract by Selection)
- [F1] Draft “AI permissions per site” UX mockups — context/https://www.anildash.com//2025/11/14/wanting-not-to-want-ai/ #fleeting #inbox
- [F2] Collect data on AI usage rates to cite — context/https://www.anildash.com//2025/11/14/wanting-not-to-want-ai/ #fleeting #inbox
- [F3] Evaluate local LLM Firefox extension security — context/https://www.anildash.com//2025/11/14/wanting-not-to-want-ai/ #fleeting #inbox
---
<!-- deep-read:sections end -->