Control over AI training is becoming easier to obtain without giving up a usable assistant, but that control says less about data retention than users might expect. Ars Technica reports that Google is separating Gemini’s history and training settings while extending retention with history disabled from 72 hours to two months. The obvious reading is a privacy upgrade; the incomplete part is treating permission to train and permission to retain as the same decision.
Google is removing a costly condition on refusing training
According to Ars Technica, Gemini users previously had to disable chat history to prevent their conversations and attachments from contributing to model training. That meant losing the ability to return to an earlier conversation.
The updated controls separate those choices. Users can retain their conversation history while declining future training use. Training remains opt-out, Ars reports, so the change removes a usability penalty without making nonparticipation the default.
That distinction matters: a privacy choice becomes more practical when exercising it no longer breaks a useful feature.
The retention change makes the privacy bargain more specific
Ars reports that disabling history will no longer mean the existing 72-hour retention period. Under the updated settings, that period becomes two months. The publication says Google describes retention as supporting responses, feedback processing and protection against misuse.
This gives the announcement its sharper consequence. A user can withhold chats from training while Google still retains data for other purposes. The setting governs an authorized use of information, rather than whether the information exists on Google’s systems at all.
Product comparisons that collapse those questions into a single privacy score will miss the actual bargain.
Returning users gain more than past contributors
The clearest beneficiaries are people who want conversational continuity without contributing future interactions to training. They no longer have to sacrifice their archive to express that preference.
The change offers less relief for information already supplied. Ars reports that the new training choice applies to future interactions and that reviewed chats already incorporated into training data will not be retroactively deleted.
For organizations evaluating consumer assistants, this makes timing consequential. A new preference cannot be treated as a remedy for earlier disclosures.
The supplied account leaves implementation questions open
The source does not establish how users can verify that a particular interaction was excluded from training, or independently assess enforcement of the separate controls. Ars says the rollout is global but may take a few weeks to reach everyone.
Reporting note: This analysis relies on Ars Technica’s account of the settings change. Sift & Signal has not independently tested the controls or verified Google’s data-handling practices.
The next test is whether the interface preserves the distinction
Watch whether the finished settings explain retention as clearly as they explain training consent. Separate toggles are useful only if users understand what each one leaves untouched.
The question to ask an AI provider is getting more precise: after I refuse training, what do you still keep?
What remains unverified
- How users can verify that particular interactions were excluded from training.
- Whether the implemented controls enforce the data-use boundaries described in the reporting.
- When the updated settings will reach every eligible account.
Sources and reporting note
- Google makes it easier to use Gemini without training AI on your data — Ars Technica reporting
Reporting note: This analysis was generated by an automated workflow using the cited public sources, source checks, and a separate AI review. Company statements and other publishers' reporting are attributed to their sources; no independent product testing or interviews were conducted by Sift & Signal.
Originally published at Sift & Signal.
Top comments (0)