Tell HN: Gemini trains on chat history, even for subscribers (personal accounts)
TL;DR Highlight
Gemini personal accounts may train on and human-review chats, even for paying subscribers
Who Should Read
A developer pasting code, keys, or customer data into a personal Gemini account at work
Core Mechanics
- No article body
- This is a Tell HN post and no body text was captured.
- Beyond the title, facts here come only from comments.
- What the title claims
- The title says Gemini trains on chat history even for personal-account subscribers.
- Settings paths and policy wording cannot be confirmed without the body.
Community Reactions
- Opinion that human review is industry-wide
- One commenter said all frontier providers monitor interactions even with activity tracking off.
- They linked NYT Wirecutter and 404 Media pieces.
- Rebuttal that the terms have reasons
- A commenter quoted terms saying reviewed chats are kept up to three years even after deletion.
- They read this as legal retention for chats that may create liability.
- Opinion that opting out of training is not enough
- A commenter cited Google's terms licensing content for operating services and building features.
- They said training exclusion is a distraction since other uses remain.
- Experience of Pixel features breaking with history off
- A user said a reminder request failed because Gemini said it needs chat history enabled.
- Opinion that only Workspace offers history without training
- A commenter said keeping history while opting out of training currently needs Gemini in Google Workspace.
- They also said Antigravity has no training opt-out.
- Opinion that the API gives confidentiality
- A commenter said cheap subscriptions are paid for with data, while the pricier API comes with confidentiality.
How to Apply
- Audit what you pasted into personal Gemini
- Spend 30 minutes today checking recent chats for keys, tokens, or customer data.
- Revoke and reissue any key you find.
- Use API or Workspace for work data
- Set a team rule that company code and customer data stay out of personal accounts.
- Move confidential tasks to an API or Workspace account.
- Mask secrets before pasting
- Replace keys, emails, and names with placeholders before sending prompts.
- Automate this with a masking script in your tooling.
Try It
# Run in terminal (no public code in source; a simple masking script to hide secrets before pasting)
import re,sys
t=sys.stdin.read()
t=re.sub(r'(?i)(api[_-]?key|token|secret)\s*[=:]\s*\S+',r'\1=<MASKED>',t)
t=re.sub(r'[\w.+-]+@[\w-]+\.[\w.-]+','<EMAIL>',t)
print(t)
# Usage: pbpaste | python mask.py | pbcopy
# Next: add your team's common secret patterns to the regexTerminology
Related Papers
Progress on provable inference for large language models
Proving 400B model outputs with zkSNARKs works, at 70-95x the GPU time of plain vLLM
You don't need long context
Compacting context to 16K–48K kept accuracy while cutting estimated costs 2–8x
The AI cost of JavaScript frameworks
Building the same app with 12 JS frameworks via agents, 5 runs each, showed up to ~2x cost differences.
Claude Code regularly sends each background agent’s transcript to its model
Disabling Claude Code's 30-second agent status summaries removed 19.7% of API requests in one team's telemetry
Cross-request KV cache exposure in the GLM-5.3 Workers AI model (155 responses)
A KV pool smaller than the context window leaked other customers' data into 155 long-prompt responses.
Microsoft Decision-1: 9B Qwen Model Launch Explained
Microsoft's 9B Qwen-based model returns only per-option probabilities for routing and classification.
Cerebras Deploys Trillion-Parameter AI Models
Cerebras fits Kimi K3 (3T) across 59 wafers in a pipeline, serving 1000-2000+ tokens per second
A practical guide to AI image generation, from diffusion to agents
Testing image APIs showed you pay per canvas, and camera-style prompts beat describing the view