Cascadity Audio
← All streams

#training-data

Everything tagged training-data, across every stream.

0

Paying for Gemini doesn't stop Google training on your chats: the only off switch also erases your history

Google's own privacy hub spells out what the Keep Activity setting does, who reviews chats, and how long reviewed chats are kept.

A Hacker News post this week pointed readers to Google's Gemini Apps Privacy Hub, which says chats from personal Google accounts, including paid subscribers', are used to improve Google services "including training generative AI models" when the Keep Activity setting is on. AI Weekly notes that Keep Activity is on by default for adults.

Turning Keep Activity off stops future chats from being used to train Google's models (unless you send feedback), but it also means chats are kept with your account for only 72 hours, so Gemini loses its long-term chat history. Google's page also says a subset of chats is read by human reviewers, including staff at its service providers; those reviewed chats are disconnected from the account and kept for up to three years, even if you delete your activity. Google asks users not to enter confidential information they wouldn't want a reviewer to see.

Temporary chats are not used to train Google's AI models, the page says. Work and school accounts fall under different terms.

Cascadic Analysis 1
UndertowConversations are a new kind of sensitive data

What hidden risk could pull against this, even if the news is good?

People tell chatbots things they wouldn't type into a search box, and Google's page says personal-account chats, paid ones included, are used to train its models when Keep Activity is on, as it is by default for adults. Some chats are read by human reviewers, including staff at service providers, and those copies are kept for up to three years even after you delete your activity. Opting out of training also cuts saved history to 72 hours, so privacy and convenience come as a package, and Google's own advice is not to enter anything confidential.

0
Rabbit Holes 1
0

Google, Microsoft, OpenAI and Anthropic plan to honor one robots.txt line for opting out of AI training

Search Engine Journal compares each company's announcement and timeline side by side, including what the shared line won't do.

Google, Microsoft and OpenAI announced on Oct. 30 that they plan to honor a shared robots.txt group for AI-training preferences, so a website can state one choice to every participating provider instead of writing rules for each company's crawler. The token, __usage_ai-training, comes from an individual IETF Internet-Draft by Anthropic's Kevin Kelley, who wrote that Anthropic has started building support, according to Search Engine Journal.

None of the companies supports it yet. Google says it aims to in early 2027 and will treat the opt-out as a Google-Extended block when a site doesn't set Google-Extended itself; Microsoft will start honoring it Feb. 1, 2027, adds a separate page-level control for AI grounding, and will stop treating its old NoArchive/NoCache tags as training signals after March 1; OpenAI says "in the coming months."

Engadget notes the line can't undo past scraping, and it doesn't block crawling or affect appearing in search results. Existing robots.txt rules aren't read as a training preference.

Rabbit Holes 1