ChatGPT Random Foreign Language Text

What started on March 23rd as bemused forum commentary has quietly stretched into a six-week pattern that's harder to dismiss as coincidence. The opening posts had the tone of people who thought they'd stumbled onto something absurd — "using ChatGPT now requires knowledge of arabic, thai, and hindi apparently" — followed almost immediately by a second user reporting Arabic characters appearing inside otherwise normal English responses. Bafflement rather than alarm, at first. Then March 25th brought a more pointed question: "Does your ChatGPT randomly throw Arabic words lately?" — the phrasing of someone who has started wondering whether they're alone in this. They weren't. A two-week quiet stretch followed, the kind that usually means either a silent fix or the internet moving on. It turned out to be neither. On April 9th the threads re-opened, two separate posts on the same day both titled "Sudden Arabic characters. New feature?" — the question mark doing a lot of work, caught somewhere between genuine curiosity and sarcasm. April 10th added a user who simply noted the model "accidentally used arabic for no reason, weird." By April 11th the tone had shifted. "Why does it keeps doing this... ChatGPT keeps adding random arabic text in answers and they still haven't fixed this lmao." The laughter is still there, but it's the laughter of someone who has given up expecting better. Then the script changed. April 29th saw a report of ChatGPT switching entirely to Russian, unprompted. May 1st added a post titled simply "What the......" — which tells you everything about the user experience and nothing about the cause. The most recent signal, dated May 5th, returns to Arabic. Whatever is happening, it hasn't stopped, and it appears to be expanding its linguistic repertoire.

If confirmed, here is what this means. A model inserting random characters from scripts it wasn't prompted to use isn't a cosmetic glitch — it points toward something misfiring in the token prediction or language-detection layer, the kind of error that erodes the basic trust a user needs to rely on a tool for real work. The shift from Arabic and Thai to Russian between late April and early May suggests this isn't a static bug tied to a single model update; it's either recurring across deployments or getting worse. For professionals using ChatGPT in client-facing contexts — drafting documents, generating code comments, writing communications — an unpredictable language switch isn't a curiosity, it's a liability. More broadly, if OpenAI is aware of this and hasn't addressed it publicly after six weeks of user reports, that silence is its own data point about how the company currently weighs forum-level feedback against product priorities.

Watch for an official acknowledgment from OpenAI, or for the reports to spread beyond community forums into verified technical analysis — either would significantly change the weight of this story.

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 →