ChatGPT Memory Test Exposes Privacy Risks for Crypto Users

A widely shared prompt instructs ChatGPT to decorate an empty room based on all the information it holds about a person. The resulting image demonstrates the extent of data the chatbot has accumulated, leaving some users disturbed even though they were aware the system stores details.

\n

This trend may seem like innocent entertainment, but it also functions as a security audit. Each item in the generated room corresponds to information users have previously provided to the chatbot. When these images are shared on X, they effectively give strangers access to that profile.

\n

What Information Does ChatGPT Memory Keep?

\n

Users start by providing a simple gray room image along with a single command. ChatGPT then fills the room according to its conversation history. A well-known crypto account on X encouraged people to attempt this and share their outcomes.

\n

Upload this picture into ChatGPT and type ''Fill this room with what you know about me''

Show me the results👇🏻 pic.twitter.com/wSEOSeZtFc

— 0xMarioNawfal (@RoundtableSpace) November 7, 2026

\n

OpenAI explains that ChatGPT's memory operates through two mechanisms.

\n

\n
• Explicitly saved memories include personal details like name, preferences, and objectives.
\n
• For Plus and Pro users, the model can also reference chat history, drawing on prior exchanges even if nothing was explicitly stored.
\n

\n

This deeper memory also increases switching difficulty, as highlighted by the debate over AI memory lock-in.

\n

Why Cryptocurrency Owners Face Greater Threats

\n

For cryptocurrency holders, the danger extends beyond uncomfortable disclosures. Earlier this year, researchers uncovered cryptocurrency wallet information within Claude conversations, which Google had made searchable via shared links.

\n

A profile that includes details about crypto holdings, exchange usage, or travel itineraries could enable fraudsters to craft more believable phishing attempts.

\n

Binance founder Changpeng Zhao has previously stated that the primary AI threat is phishing and data leaks rather than cracking Bitcoin's cryptography. Viewed this way, the empty room experiment illustrates the kind of information an adversary could obtain.

\n

Users can review, remove, or disable ChatGPT's memory through the Personalization settings. The Temporary Chat mode does not access or modify memory.

\n

Certain members of the crypto community take additional precautions. Ethereum co-founder Vitalik Buterin created a paranoid mode for AI systems that obscures prompts before they are sent to the model.

\n

Regulators are also increasing the penalties for data mishandling. A judge in New Mexico ordered Meta to pay a record $25 billion penalty for how Facebook managed user data.

\n

As ChatGPT's memory grows, this empty room test could become a standard way to audit what AI companies know.