you're sure it's not their "feature" that calling the api with empty string returns random hallucinations?

https://jarbon.medium.com/gpt-prompt-bug-94322a96c574

No, definitely not the empty string hallucination bug. These are clearly real user conversations. They start like proper replies to requests, sometimes reference the original question, and appear in different languages.

I don’t see anything here that would prevent a LLM from generating these. Right?

In one of the responses, it provided the financial analysis of a not well-known company with a non-Latin name located in a small country. I found this company; it is real and numbers in the response are real. When I asked my ChatGPT to provide a financial report for this company without using web tools, it responded: `Unfortunately, I don’t have specific financial statements for “xxx” for 2021 and 2022 in my training data, and since you’ve asked not to use web search, I can’t pull them live.`.

Do you understand what a hallucination is?

Coming up with accurate financial data that you can't get it to report outright doesn't seem like one.

Models do not possess awareness of their training data. Also you are taking at face value that it is "accurate".

I don't understand the wording

Accurate financial data?

How do we know?

What does using not-web-search not having the data have to do with the claim that private chats with the data are being leaked?

> I found this company; it is real and numbers in the response are real.

???

Which of my questions does that answer?

That the financial data is accurate?

It's an ourobos - he can't verify it's real! If he can, its online and available by search.

Therefore what are the odds that this is just the LLM doing its thing versus "a vulnerability". Seem like a pretty obvious bet.

> numbers in the response are real.

OpenAI very well may have a bug, but I'm not clear on this part. How do you know the numbers are real?

I understand you know the name is the company is real, but how do you know the numbers are real?

It's way may than anyone should need to do, but the only way I can see someone knowing this is contacting the owners is the company.

i had the exact same behavior back in 2023, it seemed like clearly leakage of user conversations - but it was just a bug with api calls in the software i was using.

https://snipboard.io/FXOkdK.jpg

There was an issue with conversation leakage, though. It involved some bug with Redis.

I felt like it was a huge deal at the time but it’s surprisingly hard to quickly google it.

It was the classic "oh no we did caching wrong" bug that many startups bump into. It didn't expose actual conversations though, only their titles: https://openai.com/index/march-20-chatgpt-outage/

ah there it is. thanks for jogging my memory. funny to think of how niche chatgpt was considered then to now.

New Touring Test unlocked! Differentiate between real and fake hallucinations.

So THAT'S what the "GT" means on all of these GPU model names!

Users should always avoid sharing sensitive data.

A lot of AI products straight up have plan text logs available for everyone at the company to view.

Which ones? Do you just mean tiny startups and side projects and the like or is this a problem that major model providers have?

It's not just about sensitive data like passwords, contracts, or IP. It's also about the personal conversations people have with ChatGPT. Some are depressed, some are dealing with bullying, others are trying to figure out how to come out to their parents. For them, this isn't just sensitive, it's life-changing if it gets leaked. It's like Meta leaking their WhatsApp messages.

I really hope they fix this bug and start taking security more seriously. Trust is everything.

maybe you should stop trusting random people on the internet making extraordinary claims without proof then?

This is going to be subject to the legal discovery process with the usual safeguards to prevent leaks; in particular, the judge will directly supervise the decision of who needs access to these logs, and if someone discloses information derived from them for an improper purpose, there's a very good chance they'll go to jail for contempt of court, which is much more stringent than you can usually expect for data privacy. You can still quite reasonably be against it, but you cannot reasonably call it "plain text logs available for everyone at the company to view".

Isn't "assume vulnerable" The only prudent thing to do here?

everything is vulnerable. the question is, has this researcher demonstrated that they have discovered and successfully exploited such a vulnerability. what exactly in this post makes you believe that this is the case?

No? Yes? Mu?

After some hemming and hawing, my most cromulent thought is, having good security posture isn't synonymous with accepting every claim you get from the firehose

Permanent NDA's? Oof. It's like their plan is to just try to force the lid down till they reach ASI or something lol

Again: NDAs are bog standard bounty terms.

The NDA part feels really murky.

It's pretty standard for bounty programs. If you don't like it, which is reasonable, do what this researcher did and just post independently.

The bug bounty world is a funny one. I remember one complaining that their bug was dismissed and fixed after they signed an NDA, no payout, nothing. Another one got $100 instead of $5,000 because the company downgraded the severity from high to low. So they ended up with little or no money, and no recognition either. Not sure if these were edge cases, but it does make you wonder how fair the process really is.

If you're dealing with large companies, a good rule of thumb is that the bounty program is incentivized to pay you out. Their internal metrics improve the more they pay; the point is to turn up interesting bugs, and the figure of merit for that is "how much did we have to spend". At a large company, a bounty that isn't paying anything out is a failure.

All bets are off with small random startups that do bug bounties because they think they're supposed to (most companies should not run bounties). But that's not OpenAI. Dave Aitel works at OpenAI. They're not trying to stiff you.

Simultaneous discovery (either with other researchers or, even more often, with internal assessments) is super common. What's more, you're not going to get any corroboration or context for them (sets up a crazy bad incentive with bounty seekers, who litigate bounty results endlessly). When you get a weird and unfair-seeming response to a bounty from a big tech company, for the sake of your own sanity (and because you'll probably be right), just assume someone internal found the bug before you did, and you reported it in the (sometimes long) window during which they were fixing it.

Interesting insights, thanks for sharing

That's an exaggeration. Most industry leaders do not require NDAs, only coordinated disclosure.

Mozilla's program, which has been around longer than most, doesn't. Google and Microsoft don't. Meta and Apple don't.

This is water carrying, intentional or not, for a terrible practice that should be shamed, so that it doesn't become standard.

My understanding is that all Bugcrowd bounties do by default.

You can shame it all you want, but you can also just publish your bugs directly. Nobody has to use the Bugcrowd platform. You don't even have to wait 45 days; I don't buy these "CERT/CC" rules.

You said it was pretty standard for bug bounty programs, and I disagreed pointing to several of the largest and longest lived bug bounty programs, none of which do that, and your response is pointing out that one particular platform does it?

Even among 3rd party platforms, of which there are several bigs, the NDAs are not a platform requirement, just an option for participating firms.

NDAs are not the norm. Don't mislead people who would otherwise get into this game with non-issues they need not worry over.

OpenAI's security team commented on the thread themselves that they believe they simply accepted the Bugcrowd defaults. I think you're trying to find a controversy that just isn't here.

well done, sounds very reasonable and following the rules.

Appreciate it. Just trying to do the right thing by both OpenAI and users here.

I see other users conversations on my Gemini dashboard, not sure who to even complain to.

Software quality is... Minimal now days.