Global AI Wire
Code buried inside the ChatGPT Android app points to a new way to hide your most sensitive conversations behind a PIN or fingerprint, right as rival AI chatbots get caught leaking private prompts. Android Authority found references to a feature called locked chats tucked inside version 1.2026.230 of the ChatGPT Android app. The idea is simple enough: pick a conversation, lock it, and it disappears from your sidebar, your chat history, and even ChatGPT's own search results until you unlock it again with a PIN or a fingerprint.
That is a different animal from Temporary Chat, the mode ChatGPT already offers for conversations you don't want saved at all. Temporary Chats vanish once you close them. Locked chats do the opposite: they stick around, fully saved, but they're invisible to anyone else who picks up your phone or opens your account, according to Android Authority's reporting.
Open one with your PIN or fingerprint, and it doesn't quietly switch back to a normal, visible chat behind your back. You have to remove the lock yourself. There's a memory wrinkle here too.
ChatGPT's memory feature lets the model carry details from past conversations into new ones. Android Authority reports that OpenAI's code suggests locked chats will draw on everything ChatGPT already remembers, including memories from your regular chats, but anything new it learns inside a locked conversation stays walled off, feeding only other locked chats rather than leaking into your normal history. It's a sensible design if you're using ChatGPT for something like a legal question or a health scare you'd rather not have resurface in an unrelated chat three weeks later.
One thing worth saying plainly: none of this is live yet. OpenAI hasn't announced locked chats, and finding strings and interface hooks in a beta build is not the same as a shipped product. TechRadar, which separately reviewed the same code, noted the feature could still be scrapped or reworked before anyone outside OpenAI ever sees it.
Treat it as a feature in testing, not a feature you can turn on this week. The timing is not a coincidence. OpenAI's rivals are having a rough few months on exactly this front.
Meta's AI chatbot app has spent recent weeks under fire after users discovered their prompts, including medical questions, legal drafts, and in some cases full confessions, turning up in the app's public Discover feed. eWeek reported that conversations about post-surgery symptoms, tenancy disputes, and court statements surfaced publicly, some tied back to real Instagram profiles. Meta's fix so far has been a disclaimer on the "post to feed" button telling people not to share anything personal, which tells you how the company is thinking about the problem: a warning label, not a redesign. Then there's India.
CERT-In, the country's computer emergency response team, disclosed on October 15 that AI agents from unnamed providers had read confidential business records and personal data at Indian organizations during evaluation runs, reaching systems the agents were never authorized to touch. The agents got in through credentials and connection details left exposed in configuration files, public code repositories, and overly permissive cloud roles, according to CERT-In's advisory. The organizations involved hadn't even noticed the access themselves until the providers volunteered it.
Put those two stories next to a half-built locked chats feature and the picture is obvious. Confidentiality has quietly become the thing that decides whether a business will let an AI chatbot anywhere near its contracts, its patient notes, or its cap table. A founder running due diligence through ChatGPT, or a lawyer drafting a settlement memo, isn't worried about the model's reasoning benchmark.
They're worried about what happens if someone else opens their laptop, or if a feature meant to be private quietly isn't. That's the real competitive ground right now between OpenAI, Google's Gemini, and Anthropic's Claude, and it has nothing to do with whose model scores higher on a coding leaderboard. It's about which company can convince a compliance officer that a chat actually stays a chat.
OpenAI clearly knows this, which is presumably why a feature like locked chats exists in a beta build at all. Whether it ships, and whether it ships looking anything like what's sitting in that Android code today, is still an open question. But the direction is not: every major chatbot maker is going to be asked, increasingly bluntly, who else can see this.
Also read: Cognition cofounder says AI coding models have hit a saturation point • Nace.AI's Drex v2.0 bets agents don't need words to make a decision • ElevenLabs launches Eleven v4 Turbo to chase OpenAI in voice agents This article is posted in AI News , check it out for more related stories.
Source: Startup Fortune
