{"id":13622,"date":"2026-10-28T10:00:00","date_gmt":"2026-10-28T10:00:00","guid":{"rendered":"https:\/\/meriaky.com\/?p=13622"},"modified":"2026-10-28T10:01:02","modified_gmt":"2026-10-28T10:01:02","slug":"ai-data-privacy","status":"publish","type":"post","link":"https:\/\/meriaky.com\/en\/blog\/ai-data-privacy\/","title":{"rendered":"AI data privacy: what your tools actually keep"},"content":{"rendered":"\n

AI data privacy depends on precise settings, not automatic protection.<\/strong> What you type into Claude can be kept for five years. Gmail’s confidential mode doesn’t stop a screenshot. And your WhatsApp backup is only end-to-end encrypted if you turn it on yourself. So every tool has its own setting for AI data privacy, tucked away in a submenu nobody stumbles onto by accident.<\/p>\n\n\n\n

Claude: the setting that keeps your chats for five years<\/h2>\n\n\n\n

In Claude, a single setting decides how long Claude keeps your chats. In August 2025, Anthropic asked every person, once, whether they agreed to let the model train on their chats. Saying yes means Claude keeps your new chats for five years. Saying no drops that to thirty days, according to Anthropic’s official announcement<\/a>. Important: Claude always keeps a thumbs up or down on a reply for five years, even if you turn the setting off.<\/p>\n\n\n\n

To check yours, open Claude and click your name in the bottom left. Then go to Settings, then Privacy, and turn off the option about helping improve the models. Thirty seconds, and that covers your next chats. This rule doesn’t apply to Team and Enterprise plans: it only targets personal accounts, including your own team’s.<\/p>\n\n\n\n

Gmail and AI data privacy: confidential mode protects less than its name suggests<\/h2>\n\n\n
\"AI<\/figure>\n\n\n\n\n

Gmail’s confidential mode, the little clock with a lock icon when you’re writing an email, blocks forwarding, copying and printing. It also lets you add an expiration date or a code sent by text. What it never does, though, is encrypt the message end-to-end. Google says so itself, in plain language: the person receiving the message can still take a photo or a screenshot of it, according to Google’s official help page<\/a>. Gmail encrypts a regular email only in transit, which is not the same thing as end-to-end encryption.<\/p>\n\n\n\n

For an actual secret, like a password or a sensitive number, email is never the right path, confidential mode or not. Say it out loud on the phone instead, or use a message that disappears after it’s read.<\/p>\n\n\n\n

WhatsApp protects your messages, not always your backup<\/h2>\n\n\n\n

WhatsApp encrypts messages end-to-end by default, and that never changes. What changes is the backup WhatsApp sends to Google Drive or iCloud. By default, that backup doesn’t have the same level of protection. WhatsApp itself describes backup encryption as an extra, optional layer, according to WhatsApp’s official announcement<\/a>. Until you turn it on, Google or Apple holds the backup. If someone ever hacks your WhatsApp account, that backup becomes readable.<\/p>\n\n\n\n

To turn it on, go to Settings, then Chats, then Chat backup, then “End-to-end encrypted backup.” Then you get three ways to protect it: a passkey, a password of your choice, or a 64-digit code. If you lose that key or password, no one, not even WhatsApp, can recover the backup.<\/p>\n\n\n\n

iPhone: a call that never rings is still a missed call<\/h2>\n\n\n\n

Less a data leak than a customer-service trap, but the same instinct to check a hidden setting. On iPhone, unknown-caller filtering can silence an incoming call without it ringing once, according to Apple’s official guide<\/a>. The call goes straight to voicemail. It’s still visible in your recent calls, but later, too late to answer right away. So if you hand out your personal number to clients, a prospect calling you for the first time is, by definition, an unknown number to your phone.<\/p>\n\n\n\n

The setting lives in Settings, then Apps, then Phone, then Silence Unknown Callers. You get three choices: never, ask why they’re calling, or full silence. The second choice still lets the call ring, with an extra step first.<\/p>\n\n\n\n

What never goes into an AI or an email<\/h2>\n\n\n\n

Whatever setting you turn on, one rule stays simple: a card number, a password or a client’s confidential number never go into an AI chat or an email, not even if you check confidential mode. These tools keep a trace, and a trace can always resurface. That’s actually the same logic behind good access management<\/a> inside your business: limiting what goes in, not just watching what comes out.<\/p>\n\n\n\n

Where to start this week with AI data privacy<\/h2>\n\n\n\n

Take ten minutes and check one single setting: Claude’s, if you use it for your business, or WhatsApp’s, if you discuss business with clients on it. One setting at a time, not all four at once.<\/p>\n\n\n\n

If you want us to go through your tools with you, reach out<\/a>.<\/p>\n\n\n\n

Frequently asked questions<\/h2>\n\n\n\n
Does Claude keep my conversations forever?<\/strong><\/strong>

No. If the training setting is turned off, your chats are kept for thirty days. If it’s turned on, it’s five years. A thumbs up or down on a reply stays kept for five years either way.<\/p> <\/div>

Does Gmail’s confidential mode stop a screenshot?<\/strong><\/strong>

No, Google says so itself: the person receiving the email can still take a photo or a screenshot of it. Confidential mode only blocks forwarding, copying and printing.<\/p> <\/div>

Why isn’t my WhatsApp backup protected the way my messages are?<\/strong><\/strong>

Because end-to-end encryption on the backup is an option you have to turn on yourself, not an automatic setting. Without it, Google or Apple holds your backup on their own servers.<\/p> <\/div> <\/div>\n\n\n\n