Many warn of the risks of AI. “AI” today typically means Large Language Models (“LLMs”) and tools that make them available, such as ChatGPT and Claude. The term can also refer to technologies such as facial recognition software and social media recommendation algorithms. You may also hear the term “Generative AI” (“GenAI”), which includes LLMs as well as image and video generation platforms like Sora and DALL-E.
Unfortunately, these tools confidently present inaccurate information as correct. “Private” conversations with ChatGPT (or other chatbots) can be leaked in data breaches or subpoenaed by law enforcement. Using AI for activism work can dangerously compromise your information security. We recommend AGAINST using AI for activism work.
AI Confidently Makes False Claims
AI tools “hallucinate,” meaning they generate inaccurate statements. If you’ve seen AI images with made-up words or misshapen hands, it’s a similar concept. One example is inventing fictional cars for someone trying to research which car to buy. OpenAI has published a research paper demonstrating that these hallucinations cannot be fixed. It’s easy to be fooled by hallucinations, since AI often presents them in an authoritative tone.
AI-produced information may be correct, but when accuracy matters you should verify information directly with trustworthy sources, just like fact checking a claim from a friend or from someone on the Internet. You cannot trust citations from the AI, as it hallucinates those as well.
AI Isn’t Really Private
Conversations with AI give the impression of privacy, but many of these chats have been made publicly visible in recent history in various incidents. ChatGPT, Meta (Facebook), and xAI (aka Twitter, aka Grok) have all created “features” that led to users making their chats publicly readable without realizing. AI companies are also selling customers’ data. You can safely assume that any conversations you’ve had with AI since July 2025 have been captured and sold to third parties.
AI + Activism: The Wrong Kind of Trouble
Any use of AI tools can also be accessed by the police. Chat logs can be subpoenaed by courts even if you delete them.
Recommendations
Don’t use them for activism work. Turn them off in any way you can.
Easy Mode
We still recommend using DuckDuckGo or another Google alternative search engine. If you need to use Google, you can use udm14.com to default to the Web search and bypass some of Google’s forced AI bundling.
Unfortunately, as AI adoption increases, avoiding it becomes more and more difficult. There are small steps you can take in your day-to-day life to minimize your use of it. We recommend using DuckDuckGo as a general search engine (and opting out of its AI features). Other search engines like Startpage also offer privacy-focused searching with limited or no AI features.
Using a privacy-focused browser such as Firefox can also minimize your exposure to AI features. Mozilla introduced AI controls to allow users to limit or entirely disable current & future GenAI features in the browser.
Chrome will collect as much of your private information as it can. Don’t use Chrome.
Level Up!
AI tools are evolving more rapidly than we can update this website, so check these publications and other trustworthy sources for current instructions for turning off LLM tools.
- How to turn off LLM features, by Consumer Reports (October 3, 2025)
- Turning off AI features in Gmail, Photos, and Chrome (February 24, 2026)
- Turn off Apple Intelligence (March 4, 2026)
- Turn off Gemini on Android (July 11, 2025)
Hard mode
If you really really really want to keep using an LLM, this article gives some options for more privacy-friendly options. It is targeted at people who are ready to invest in their own personal setup to use a private LLM.
Don’t panic
Like it or not, AI has become highly integrated into modern life, and turning it off is not always clear or easy. Start where you are and be kind to yourself.
Note: AI and Surveillance
Is the government using AI to supercharge mass surveillance?
Local, state, and federal governments in the US are indeed using LLM-based tools for law enforcement, surveillance, and associated tasks. These tools and others based on similar technologies are expanding mass surveillance capabilities beyond what previously existed.
Be cautious of excessive alarmism in news reporting and analysis, however. In articles like this one, be mindful of what the writer says could or might happen as opposed to what is currently happening.