On November 7, 2026, the crypto account 0xMarioNawfal urged followers to ask ChatGPT to “Fill this room with what you know about me” and publish the resulting image. The trend, covered by BeInCrypto on November 8, highlights how personal details voluntarily shared with an AI chatbot could give scammers material for more convincing crypto phishing messages.

The prompt does not show a wallet exploit or an unauthorized break-in to ChatGPT. It turns remembered conversation context into a visual profile, and the privacy concern begins when users place that profile in front of strangers on X.

What Happened: A ChatGPT Room Prompt Turned Personal Context Into Shareable Images

What did the prompt ask users to do?

The instruction used an image of an empty gray room. Users were asked to upload it to ChatGPT and type, “Fill this room with what you know about me.” The chatbot then furnished the room with objects representing details drawn from earlier conversations.

The result is designed to feel personal because the objects reflect information a user previously typed into the service. That can make the image entertaining, but it can also expose clues about a person’s interests, plans and financial activity when the image is posted publicly.

When did the sharing invitation appear?

The post reproduced in the report is dated November 7, 2026. It encouraged followers to “Show me the results” and was attributed to 0xMarioNawfal, using the handle @RoundtableSpace.

The privacy-risk article appeared at 01:18 UTC on November 8. Those dates establish the sequence documented around the prompt, but they do not establish when the wider trend began or how many people took part.

What changes when users post the image?

Generating the image and publishing it are separate actions. The first uses personal conversation context to create a response; the second makes the resulting profile available to people outside the user’s ChatGPT account.

That distinction matters for crypto holders. A stranger does not need access to the original conversations if a public image reveals enough personal context to suggest what the user owns, where they trade or when they may be away from home.

Who Is Involved: OpenAI, the Promoting Account and Crypto Security Voices

Who encouraged followers to share results?

The account promoting the prompt was 0xMarioNawfal, identified in the embedded post by the handle @RoundtableSpace. The supplied reporting does not describe the account as an OpenAI campaign or establish an affiliation between it and the company.

The account’s role was to popularize the instruction and invite users to display the generated images. The exposure risk described here comes from voluntary publication, not from a claim that the account accessed users’ private chats.

What is OpenAI’s role?

OpenAI provides ChatGPT and the memory controls described in the report. Its system can use information from earlier interactions in the ways outlined below, which is why the room prompt can produce a response that feels tailored to a particular user.

OpenAI has also been making separate public commitments about its wider AI infrastructure. Decrypt described a November 5 community compact covering electricity, water and reporting at future campuses, but that initiative is separate from ChatGPT memory and the viral prompt.

What do crypto figures add to the discussion?

The report cites Binance founder Changpeng Zhao’s view that AI’s immediate danger to crypto is phishing and leaked data rather than an attack on Bitcoin’s cryptography. That is his assessment of the threat, not a documented reaction to this particular room prompt.

It also mentions Ethereum co-founder Vitalik Buterin’s “paranoid mode,” which masks prompts before they reach an AI model. The cited approach is presented as a privacy measure; the supplied material does not provide setup instructions or a guarantee that it prevents every form of disclosure.

Background: ChatGPT Memory Can Draw on More Than One Source

What can saved memories contain?

According to OpenAI as summarized in the report, saved memories can include a user’s name, preferences and goals. These details are retained as information the model can use to tailor later interactions.

That list is not described as a complete inventory of everything a user has ever typed. The room image should therefore be read as a visualization of personal context, not as a certified export of all information held by OpenAI.

How does chat-history reference differ?

The report describes a second mechanism: chat-history reference. It says Plus and Pro subscribers can receive responses that draw on earlier conversations even when a detail was not saved as a separate memory.

This distinction helps explain why a user may see personal material represented in the generated room without recognizing every object as a manually saved memory. The source describes the feature availability in those terms, so the Plus and Pro qualification should not be extended to other plans.

What earlier incident illustrates the sharing risk?

Earlier in 2026, developers found crypto wallet data in Claude conversations that Google had indexed through shared links, the report says. That example involved publicly shared Claude chats and search discovery.

It is not the same mechanism as the ChatGPT room trend. The earlier case concerned shared links becoming discoverable, while the current concern is that users may publish a generated image containing clues from their own conversations.

Why It Matters: Personal Context Can Make Crypto Phishing More Convincing

Which details could help a scammer?

The report specifically identifies holdings, exchanges and travel plans as details that could help scammers write more persuasive messages. A message that references a user’s likely exchange, a recent trip or an apparent interest in a particular asset may look more credible than a generic request.

The risk is conditional. A generated image may contain useful clues, but it does not establish that every user holds crypto, uses a particular exchange or has exposed a private key.

Why are crypto holders an attractive target for context-driven scams?

Crypto activity often involves accounts, wallet access and urgent transaction decisions. Personal context can help a scammer imitate a familiar situation, such as an exchange alert, a travel-related request or a message timed around the user’s apparent plans.

The supplied reporting does not describe theft caused by this trend, a compromised wallet or a confirmed phishing campaign tied to the images. Its point is narrower: voluntarily disclosed context can improve the quality of a future social-engineering attempt.

Why is public personalization different from a system breach?

The image trend shows a user asking the model to generate a personalized response and then choosing whether to share it. That is different from an attacker bypassing OpenAI’s systems or extracting private keys from the chatbot.

The practical concern is the information pathway. A user may reveal more than intended through a harmless-looking image, and a stranger can combine those clues with information from other public accounts.

Two Memory Mechanisms and the November 7–8 Timeline

What do the two mechanisms show?

Saved memories retain selected personal details such as names, preferences and goals. Chat-history reference, as described for Plus and Pro subscribers, can draw on earlier chats even when a detail was not separately saved.

The prompt relies on the model’s ability to use personal context, but the supplied material does not quantify how much information any one response uses. The objects in an image should not be treated as a full record of the user’s account history.

What does the November sequence establish?

The documented post appeared on November 7, 2026, and the privacy-risk report followed on November 8 at 01:18 UTC. This places the invitation to share results before the article describing the crypto-phishing concern.

It does not establish the origin of the wider trend, its total reach or whether any user lost funds after posting an image. Those figures are not part of the reported account.

Which subscriber tiers are named?

Plus and Pro are the only subscription tiers specifically associated with chat-history reference in the supplied report. The same account distinguishes that feature from saved memories, which can contain a user’s name, preferences and goals.

That distinction is useful when reviewing settings, but it should not be turned into a broader claim about every ChatGPT plan or every type of data the service may process.

Risks and Evidence Limits: A Phishing Pathway, Not a Wallet Exploit

What could go wrong after an image is shared?

An attacker could use apparent knowledge of holdings, exchanges or travel plans to make a phishing message sound plausible. The more personal the context, the easier it may be to frame a request as something the target expects.

Users who publish the image also place the clues in a format that can be copied, archived or combined with other public information. The report does not provide an attack success rate, a loss figure or a confirmed victim tied to the trend.

What exposure does the report actually describe?

The exposure described is voluntary publication of an AI-generated representation of information from prior conversations. It is not a claim that ChatGPT automatically posts memories, that the prompt extracts seed phrases or that OpenAI’s memory system has been breached.

The earlier Claude example should remain separate. In that case, the reported discovery route involved shared links indexed by Google; it was not presented as a demonstrated compromise of ChatGPT memory.

What should readers make of the image’s contents?

The room is a model-generated interpretation of prior user information. An object that appears in the scene may reflect a conversation detail, but the image is not a verified list of assets, accounts or private credentials.

That limitation does not remove the privacy issue. Even partial clues can be useful if they reveal a user’s interests, routines or relationship with crypto services.

What Users Can Do Next: Review Memory Controls Before Sharing

Where can users manage ChatGPT memory?

The report says users can view, delete or switch off memory in ChatGPT’s Personalization settings. Reviewing that area gives users a way to inspect the memory feature before asking the model to build a public-facing profile.

The supplied material does not provide additional interface steps or claim that changing one setting deletes every category of information associated with an account. Readers should keep the action limited to the memory controls described.

How does Temporary Chat handle memory?

According to the report, Temporary Chat neither reads nor updates memory. That makes it a separate option for conversations a user does not want used in the saved-memory or chat-history context described above.

The statement concerns reading and updating memory. It does not support broader claims about all forms of data retention, account records or model-training policies.

What privacy approach does the report mention?

The report says Vitalik Buterin built a “paranoid mode” that masks prompts before they reach a model. It presents the approach as a way to reduce the amount of directly identifying information sent to an AI system.

No setup guide or security guarantee is supplied. The relevant takeaway is that prompt minimization can be considered alongside the built-in memory controls, especially before a user publishes an AI-generated profile.

FAQ

What kinds of details can ChatGPT saved memories hold?

The report says saved memories can include a user’s name, preferences and goals. It does not describe that list as a complete record of every conversation or every piece of information OpenAI may process.

Which subscriptions are associated with chat-history reference?

Plus and Pro are the tiers associated with chat-history reference in the supplied account. That feature can draw on earlier conversations even when a detail was not saved as a separate memory.

Where can users manage ChatGPT memory?

Users can view, delete or switch off memory through ChatGPT’s Personalization settings, according to the report. The supplied material does not provide further navigation steps.

Does Temporary Chat read or update saved memory?

The report says Temporary Chat neither reads nor updates memory. That statement is limited to the memory behavior described and does not address every form of account data handling.

Which crypto details could help personalize phishing?

Holdings, exchange use and travel plans are the examples identified in the report. Those details could help a scammer make a message appear more relevant, although the article does not document theft linked to this prompt.

How did the earlier Claude conversations become discoverable?

Developers found crypto wallet data in Claude conversations that had been shared through links and indexed by Google, according to the report. That was a separate shared-link exposure, not a demonstrated ChatGPT memory breach.

This article is for information only and is not financial advice.

Related Articles

MetaMask Gives AI Bots Self-Custody on Ethereum: Is Agent Wallet the Future of Crypto?

MetaMask, one of the most popular crypto wallets, is integrating AI agents into its product. We...

Justin Sun Issues Bitcoin Quantum Warning as TRON Targets Quantum-Resistant Upgrade

TRON founder Justin Sun revived the discussion of whether the largest blockchains would be...

AI SaaS Platforms and Tools for US Businesses in 2026

US teams now use AI SaaS to automate defined work without maintaining every model, server, and...

Curve Founder Says Claude Mythos May Hit OpSec Before DeFi

Claude Mythos may pose more immediate risk to multisig key security than to DeFi, Curve Finance...

X Sues Bitcoin Influencers Over Alleged £207K Creator Revenue Scheme

In the lawsuit, submitted in England, X alleged that two identified defendants, along with several...

ZetaChain Halts Cross-Chain Operations After 300K Exploit

On April 27, layer-1 blockchain ZetaChain suffered an attack on its GatewayEVM smart...