Sam Altman says OpenAI has fixed a “significant issue” in ChatGPT after a bug in an open-source library let some users see titles of other users' chat history
ChatGPT-owner OpenAI said on Wednesday it had fixed a bug that caused a “significant issue” of a small set of users …
ReutersAkash Sriram
Context & Ripple Effects
This closes out the week's ChatGPT reliability arc: after the March 20 incident forced OpenAI to temporarily shut ChatGPT down while some histories went missing, the company has now traced the leak to a bug in a Redis client open-source library — the same root cause behind the exposure of personal information belonging to 1.2% of ChatGPT Plus subscribers, per OpenAI's own post-mortem.
Altman's confirmation that the fix is live matters because it is the third such episode in the corpus — from the November 90-minute API-and-ChatGPT outage to this privacy leak — and each one lands on a product that enterprises and paying subscribers are being asked to treat as dependable infrastructure.
First-order effects
Affected users get closure but not full restoration: the fix stops cross-account title leakage, yet some users' chat histories from the shutdown remain missing, and the small set of ChatGPT Plus subscribers whose payment and contact details were exposed are already identified.
Sam Altman absorbs the reputational cost personally, acknowledging a 'significant issue' days before OpenAI pitches enterprise adoption.
Second-order effects
The root cause sitting in a third-party open-source Redis client puts dependency auditing on every AI vendor's checklist — competitors pitching enterprise contracts will be pressed on their own shared-infrastructure failure modes.
A pattern of repeat incidents (March leak, November API outage) raises the insurance-like premium on OpenAI's reliability commitments, giving rivals a concrete sales wedge against ChatGPT for uptime-sensitive workloads.
Third-order effects
If consumer chat assistants keep functioning as de facto critical infrastructure, incidents like this normalize formal operational-assurance practices — post-mortems, dependency disclosure, breach notification cadence — that currently exist only ad hoc across AI labs.
Privacy engineering for chat products shifts upstream: cross-tenant data isolation in caching layers becomes a procurement requirement rather than an implementation detail, reshaping how labs architect shared infrastructure.
The trend: Consumer AI services are accumulating an outage-and-leak record fast enough that reliability and data-isolation failures are becoming a standing governance discipline for AI labs rather than one-off fixes.
we had a significant issue in ChatGPT due to a bug in an open source library, for which a fix has now been released and we have just finished validating. a small percentage of users were able to see the titles of other users' conversation history. we feel awful about this.
This is a significant privacy breach. People are encouraged to use ChatGPT as a personal tool, so that means everything from sensitive work tasks to health questions have been exposed to other users. https://twitter.com/...
unfortunately, users will not be able to access their chat history from monday 1 am PDT until monday 10 am PDT. we will follow up with a technical postmortem.
Those of us who've designed, deployed and maintained large scale systems are shaking our heads as the mythology of ‘AI’ meets the reality of enterprise IT and software intensive systems engineering. Oh and what library? https://twitter.com/...
Operating at scale is always hard. Measure companies by how well they handle problems, not just by whether there are problems inn the first place. https://twitter.com/...
Right on, let's blame something you got for free with no warranty. Reminds me of that time people at work discovered Redis response pipelining is usually but not always FIFO (their rando unaudited scala client worked... usually ;). https://twitter.com/...
These lapses makes me very nervous about research designs that have human subjects interact with these things. Researchers need (better) control/information/oversight of these proprietary systems. https://twitter.com/...