Google separates Gemini app chat history from AI training, adding a distinct opt-out toggle for training
Edited by Elliot Marsh, Editor · written with AI from the linked sources (how)
Google is rolling out separate toggles in the Gemini app for model training and chat history. Users can now stop their chats from being used for AI training without turning off history, which was previously the only way to opt out. Training remains opt-out by default, and the toggle is under Activity in Gemini settings. It applies only to future interactions. Data already used for training is not removed. Google may still retain data for up to three years for other purposes. For users who disable history, the retention window rises from 72 hours to two months. The rollout is global and may take a few weeks.
Entities: Google, Gemini, Ars Technica, Ryan Whitwam
0 primary
What happened
Google is rolling out separate settings in the Gemini app for model training and chat history, according to Ars Technica. Until now, the only way to stop chats being used for training was to switch off history. The new training toggle sits under Activity in Gemini settings and applies to future chats only. Data already used for training is not removed. Training stays on by default. Google may still keep data for up to three years for other purposes, and for users who turn history off, the retention window rises from 72 hours to two months. The rollout is global and may take a few weeks.
Why it matters
Consumers can now keep chat history, including picking up old conversations, without their chats feeding model training. That is a real but modest gain. Because training is still on by default, most people will not change anything, so the effect on Google's data intake is probably small unless users know the toggle exists. The longer retention for people who disable history (72 hours to two months) partly cancels the privacy benefit. Regulators and competitors may treat the split as a reference point for what a reasonable opt-out looks like, but nothing in the report shows that happening.
What is noise
Calling it a "welcome privacy improvement" overstates things: it is a more granular opt-out, not an opt-in, and it does not undo past training. The report cites no Google primary document, so the details, including the three-year and two-month figures, rest on secondary reporting that appears to relay Google's statements. We also have no data on how many users will find or use the toggle.
Watch next
- 01Google's own help or policy page: does it confirm the three-year and two-month retention figures and the toggle location, and does the wording match the reporting?
- 02Rollout progress over the next few weeks: whether the toggle actually appears for all regions, including the UK and EU, and whether any regulator (ICO, EU data protection authorities) comments on the opt-out default.
- 03Whether rivals such as OpenAI, Anthropic or Microsoft change their own training and history controls, and whether Google later moves training to opt-in or discloses how many users switch it off.
Coverage
2 storiesNewsletter
Get the no-hype AI news straight to your inbox
Weekly. No spam, unsubscribe any time. Privacy
More regulation signals
Full feed →- NYT-led publishers file summary judgment brief citing internal OpenAI/Microsoft messages calling AI training "astonishing theft" and admitting chatbots substitute for journalism18 Sept 202686
- US Commerce Department issues interim rule requiring licenses for AI chip and server exports to data centers that could provide compute to China and other restricted countries16 Nov 202684
- Anthropic launches Nasdaq IPO roadshow, pricing 825 million shares at $70-$78 for up to $90.4 billion total raise at up to $1.02 trillion valuation10 Nov 202680
- Anthropic files public S-1 with SEC for Nasdaq IPO under ticker "ANT"26 Oct 202680