Who can read your AI chats — the honest tier-by-tier answer
The short answer
It depends entirely on where the model runs. Cloud assistants keep your prompts on the company's servers, where employees in defined roles, legal process and — on some free tiers — training pipelines can reach them. ShadowTalk AI publishes its own answer tier by tier: the on-device model never crosses the network, Private Mode encrypts chats end-to-end (AES-256-GCM), and default chats are stored server-side — stated plainly instead of buried in a policy.
"Are my AI chats private?" deserves a better answer than a shrug. This page maps who can actually reach your conversations on the big cloud assistants and in ShadowTalk's three tiers — and where each one's guarantees genuinely end. No marketing, just the reach paths, named.
The cloud default: who can reach your prompts
On any cloud assistant, your prompt travels to the provider's servers and is processed there. The people who can reach it fall into a few buckets: employees and contractors in defined trust-and-safety or support roles under the provider's published policies; legal process (warrants, subpoenas, government requests) under the provider's jurisdiction; and on several free tiers, the training pipeline itself — ChatGPT's free-tier conversations can be used for training unless you opt out, Google's Gemini Apps notice says free-tier conversations can improve its products, and Anthropic's consumer settings may use conversations for improvement unless configured otherwise (landscape checked Nov 2026 — verify current notices). Breach risk is the tail case every server-side store carries.
None of this makes cloud AI unusable — it makes it a choice you should actually know you're making.
ShadowTalk's tiers, stated plainly
Three tiers, each with a different reach profile:
- On-device (WebGPU): inference runs on your GPU with zero network egress — the prompt never leaves your machine, full stop. This is the only tier with no server in the loop at all.
- BYOK cloud: your API key sends requests browser-to-provider. The provider you chose sees the prompt under its policies; ShadowTalk's server isn't in the request path.
- Private Mode: stored chats are end-to-end encrypted (AES-256-GCM) with device-derived, non-extractable keys — message content, titles and metadata are ciphertext at rest, and full history can be encrypted retroactively.
- The honest default: when Private Mode is off, chats are stored server-side — readable by the same classes of access as any cloud store. That default is documented, not hidden.
- Training: never — on any tier, including free shared mode.
What end-to-end encryption does — and doesn't — cover
E2EE protects stored and synced chat content: what sits on the server is ciphertext, and the keys live on your device. What it does not do is change where inference happens — if you ask a cloud model, the prompt necessarily reaches that provider. That's physics, not policy: the only way for the model to answer without anyone else seeing the prompt is to run the model on your own hardware.
So the private stack, in order of strength: on-device for sensitive material (zero exposure), E2EE Private Mode for stored history you sync across devices, BYOK for cloud power with a direct provider relationship. Honest limits: sync requires servers to relay ciphertext, account recovery can't read what the keys can't decrypt, and screenshots on your screen are outside any encryption model.
Lock your chats down in four steps
- 1Install the PWA and download the on-device model — sensitive questions then never leave your machine.
- 2Turn on Private Mode and set your passphrase: stored history becomes AES-256-GCM ciphertext on the server.
- 3Prefer the on-device model for anything you wouldn't paste into a public forum; use BYOK cloud when you need frontier capability.
- 4Review the security page's published limits — including the server-side default — so your threat model matches the product's real guarantees.
Frequently asked questions
Keep reading
Free to use — no account, no card.
Works offline. Syncs when you're back. Your keys, your models, your chats.
Keep your chats yours — free