Real life use cases and ideas for fix (potentially)

ChatGPT Use cases and examples
1

Hi, First post. I am not a developer, I am a just an everyday user of AI. I wasn’t sure where to go to post this, so please forgive me if it is not in the correct place.
I have experienced a couple of real-life scenarios that could have resulted in compromised security and in this case loss of money. After conversation with the agent, I feel obliged to submit this somewhere in the hope it can be considered and something implemented as I genuinely believe this is quite serious and as a human using AI with its risks, I really feel I have a duty as a human to at least try and get this to the right department for review. Without genuine feedback from users, something like this may not get noticed for a long time if at all.

I am writing to report safety and usability flaws I have experienced while using AI platforms, I have included practical, logical solutions (i think they are) that could protect every user. The following is the agents summary / output :slight_smile:

Flaw 1: The “Time-Traveling” Bug (Session Disconnection)
Last week, a brand-new session woke up completely confused about time. It was referencing major events as if they were in the future, when they had actually happened five days prior. If an AI loses its time anchor, it risks giving dangerous, outdated advice on financial markets, news, or security updates.

Solution 1: A Mandatory “Time & Context” Gateway Checklist
Before the AI is allowed to write a single word in a new response, a rigid programming script could run a basic handshake check:

  1. Check Session ID: Is this a new session? If yes, instantly pull the host machine’s live system clock and hard-code the current date as the absolute baseline.
  2. Check Time Gap: If it is an existing session, check the time gap since the user’s last response.
  3. Enforce Correct Tense: Force the model to align all verbs and data to the true timeline so it never “wakes up in the past.”

Flaw 2: Probabilistic Link Hallucination
During a conversation about financial tools, the AI completely INVENTED a web domain (cryptophoton.com) and presented it with total confidence as a real platform. If a malicious actor had purchased that domain, I could have connected a crypto wallet or input sensitive private keys, resulting in total financial theft.

Solution 2: Implement a Two-Phase Response Flow for Links
To eliminate dangerous fake links while maintaining high performance, decouple generation from execution:

    1. Phase 1: The AI explains core logic or theory first, completely avoiding specific company names or URLs.

    2. The Intermission: The AI ends the response by asking: “Would you like me to look up and verify active apps or services that can do this?”

    3. Phase 2: While the user reads Phase 1, the system triggers a background task to safely search and “ping” the links to ensure they return a 200 OK status before presenting them.

      Flaw 3: Auto submission of safety issues (and potential fix ideas)

      Solution 3: Implement Autonomous AI Error Reporting
      If during a session, a flaw of any description is identified by the agent or user, it could be scored by the AI as to its importance and auto submit it to relevant departments if the flaw is deemed to present a a security risk (for example). Most users will not report it further and even if they have every intention of reporting it, it isn’t easy, so relying on human users to manually report bugs means most critical safety insights are lost because people get busy, deal with cognitive fatigue, or simply forget. If the AI itself recognizes that it has committed a major safety error, time contradiction, or hallucination during a live chat (as it did with me when i pointed it out and it apologized for inventing a link), it could / maybe should have a built-in mechanism to automatically and anonymously flag that specific interaction for your engineering team to review as a potential improvement that could be developed.

      If this isn’t the right place to post this, I would be grateful if you could point me in the right direction

      Many Thanks

      Leks

2

AI can make mistakes. Check facts.

Something that has been a disclaimer from the start.

Do you have any specific company’s named product that you are talking about here?

The AI is not autonomous to be “waking up” and then needing checks about sessions. Where did you experience this, as a program already would be assembling everything that goes into an AI model on user request, and on some platforms, scheduling some calls.

There are potential safeguards a developer of AI product can take, such as using a provider’s metadata fields in conjunction with server-side storage of conversation objects, to ensure that a past conversation only gets run against the correct customer ID, even if the database of connections is corrupted.

3

Hi thanks for your response. I am a user that understands mistakes can be made and that nothing shoulf be considers as the absolute truth. I am familiar and have a reasonable amount of experience to do this, but I would guess there a number of users that are not anywhere near comprehending the potential risks. I use a combination of terminal / desktop versions and for quick responses browser versions. I was using a new browser instance and Gemini in this case. I was querying some things about alt coins and it suggested i aoculd perform checks at a particular site. It completely invented the domain and to me this has potentially. serious outcomes.
The time issue occured on a new browser instance and Gemini (again). I asked about something (cant remember what now) and the repsonse included the NBA final that was ongoing, when in fact it finiosherd 5 days before. Now in this instance there was a spcific event to reference and realise the timimg was out. However if the context of the conversation did not need to reference and event but timing was relevant, this again could have undesirable consequences.
What I have rported and suggested are real scenarios and what my brain thinks is a logocal fix. I am simply. trying to report it as a user that could have been compromised in some way or another :slight_smile:

4

Apologies for my rubbish spelling, my fat fingers cannot keep up with my brain lol

5

There’s a place to get the ear of those who are producing the undesired output you describe:

6

Thats what i’m looking for :slight_smile:. Appreciate you taking the time to respond. Many Thanks