Chats getting erased?

What began on April 12th as two oddly specific complaints has, over six weeks, grown into something that resists being dismissed as coincidence. The opening signals split cleanly: users reporting that ChatGPT was injecting stray words of Arabic, Hindi, and outright gibberish into otherwise coherent English responses, and separately, that code outputs were being truncated at square brackets with no explanation offered. Two bugs, possibly unrelated, possibly not. Then came April 23rd — a post titled simply "What… happened?!?" with a distressed emoji, the kind of title someone writes when they open a tool they depend on and find it unrecognisable. A day later, another user discovered that ChatGPT could no longer read its own release notes. By April 26th, the complaints had shifted from behaviour to data: images, gone. Not corrupted, not inaccessible — gone. May arrived and the pattern deepened. On the 10th, a user reported losing an entire branch of conversation, describing it as losing "all of my hard work." That same day, another user found that uploaded code files were being silently ignored — present in the interface, invisible to the model. Then, on May 16th, the signal that landed hardest: "ChatGPT just murdered my workflow." Not a confused bug report. A eulogy. Six weeks, ten distinct complaints, a trajectory that moves from odd output behaviour toward something more unsettling — the quiet erasure of the work people brought with them.

If confirmed, here is what this means. The implications divide into two separate problems, and the second one is considerably worse than the first. Garbled outputs and truncated code are quality issues — frustrating, repairable, the kind of thing a model update can address. But the loss of conversation branches, disappearing images, and code files the model won't acknowledge strikes at something foundational: the assumption that ChatGPT is a reliable place to store and continue work. Professionals who have built research threads, iterative coding sessions, or creative projects inside ChatGPT are discovering that persistence cannot be taken for granted. The second-order effect is trust erosion at exactly the wrong moment — OpenAI is pushing hard toward agentic, long-horizon use cases that depend entirely on users believing their context will survive. If the floor keeps disappearing, nobody builds upstairs. Competitors who can credibly promise memory stability — whether through local storage, explicit versioning, or transparent sync — have a specific, concrete argument to make right now.

Watch for any official incident report or post-mortem from OpenAI acknowledging data or context loss, and pay equal attention to whether the May 16th "workflow" complaint begins attracting corroboration — if that thread scales, this moves from background noise to a documented reliability failure.

NewsHive monitors these sources continuously. All signal titles above link to the original reporting.

Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →