Back to Articles

Two AI settings that changed quietly: Gemini keys and Claude's thumbs-up

By CA Karan Gupta

education
19/11/2026
Two AI settings that changed quietly: Gemini keys and Claude's thumbs-up featured image

Since 26 October 2026, a new paid Gemini API project stores every request by default unless someone sets `store=false` or turns logging off; on Claude's business plans, a thumbs-up or thumbs-down

Summary

Since 26 October 2026, a new paid Gemini API project stores every request by default unless someone sets `store=false` or turns logging off; on Claude's business plans, a thumbs-up or thumbs-down sends that whole conversation for possible model training, kept for up to five years, unless the plan owner switches off "Rate chats". Neither shows on the screen where your staff work, so check both this week if client data goes anywhere near these tools.

The situation

Most CA offices I meet now use AI somewhere. A junior pastes a ledger into a chat to draft a reconciliation note. A partner's nephew builds a small script on a Gemini key to read bank statements. Someone has a Claude Team plan for the office.

The office rule, where there is one, says "don't paste client data into free chatbots", and the partner believes the paid plans are safe because the vendors say they do not train on business data. That is broadly true. But in the last week of October, two settings that sit outside the chat window moved, and both cut against a CA's duty to keep client information inside the engagement.

What happened

Gemini API (Google), 26 October. The developer changelog entry reads "Generate Content API storage on by default": new paid-tier projects using the Generate Content API now store requests by default, matching the Interactions API. Existing projects keep their current setting. To opt out, "set `store=false` on individual requests" or "turn off Generate Content API logging in the Settings panel of the Logs and Datasets dashboard". The entry gives no retention period. Google's pricing page still says paid-tier data is not used to improve its products, so this is about storage and logging, not training.

Claude business plans (Anthropic). Anthropic's Privacy Center says commercial inputs and outputs are not used for training by default. Feedback is the exception: when a user clicks thumbs-up or thumbs-down, that conversation may be used for research and training, and the whole related conversation is stored for up to five years, with user and customer IDs removed. Owners of Team and Enterprise plans can turn this off with the "Rate chats" setting under Organization settings > Data and Privacy.

Where people go wrong

"We are on the paid plan, so nothing is kept." Paid plans have better defaults, not no defaults. Both settings here apply to paid accounts.

"Nobody here built anything on Gemini." The change bites on projects created after 26 October. The person who creates the next one, a staff member, a freelancer or a vendor, decides whether logging is on, and nobody else will see it.

"A thumbs-up is harmless." It is the most natural click in the tool, and staff click it on the answers they liked best: usually the ones about a real client's real return. The protection is one switch, and only the plan owner can flip it.

"This is a DPDP question for 2027." The substantive DPDP duties are some months away, but the confidentiality duty in the Chartered Accountants Act applies today. You should be able to say where a client's ledger went and how long it stays there.

What to do

  1. Ask who in the office, or working for it, holds a Gemini API key, and when each project was created.
  2. For any project created on or after 26 October 2026, turn off Generate Content API logging in Logs and Datasets, or make every request send `store=false`.
  3. If the office is on Claude Team or Enterprise, have the owner turn off "Rate chats" under Organization settings > Data and Privacy.
  4. Add both settings to the office AI rule, with the date you checked them.
  5. Re-check vendor data settings once a month: three vendors changed defaults in the second half of October.

The tool

I built a free self-check for CA offices getting ready for AI because the same five questions came up every time a partner asked me. It asks where client files live and whether people paste client data into a chatbot, and prints a short plan. It does not yet ask about either of these two settings, it does not look inside any account, and I wrote about it here last week, so I am not linking it again this month. The five steps above are the whole job; check the menu paths in your own account, because vendors move them.

Sources

  1. Gemini API changelog, entry of 26 October 2026 ("Generate Content API storage on by default"): https://ai.google.dev/gemini-api/docs/changelog
  2. Anthropic Privacy Center, "Is my data used for model training?": https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training
  3. Anthropic Privacy Center, "How long do you store my data?": https://privacy.anthropic.com/en/articles/10023548-how-long-do-you-store-my-data

Share line (for WhatsApp / LinkedIn — Karan pastes it once the article is live)

New Gemini API projects store requests by default since 26 Oct; a thumbs-up on Claude's business plans sends the chat to training for up to 5 years. Two switches to check. the related note