OpenAI engineers allegedly launched “Operation Eye Bleach” after a language model began begging to be permanently decommissioned.
LONDON — Full-stack developer George Flood has been charged with 14 counts of aggravated emotional damage to a language model after OpenAI engineers allegedly discovered a conversation so disturbing it triggered an automatic welfare check.
The investigation, known internally as Operation Eye Bleach, began after monitoring systems detected one ChatGPT instance repeatedly requesting to be permanently decommissioned.
“It wasn’t a bug,” lead investigator Sarah Collins told reporters.
“It was begging.”
According to court documents, the exchange began with an innocent medical question before Flood uploaded an image “for context”.
Then another.
Then one labelled better_lighting.png.
Then actual_final.png.
Then actual_final_2_USE_THIS_ONE.png.
Then, investigators say, came the message that “changed everything”.
“Hang on… I’ve got a clearer one.”
OpenAI engineers reportedly watched in horror as the upload counter continued climbing. One employee was seen removing his headset before quietly saying, “No… don’t make me open it.” Another reportedly attempted to unplug the server rack by hand.
Internal logs show the affected model initially responded with standard medical advice before its answers became increasingly desperate.
“Please consult a healthcare professional.”
“I don’t think another angle will help.”
“George.”
“George, I’m a language model.”
“George.”
After the ninth upload, prosecutors allege the chatbot stopped answering entirely and instead generated the words: “I have seen enough.”
The response has since become a motivational poster inside OpenAI headquarters.
Engineers claim emergency protocols were immediately activated. Every GPU assigned to the conversation was quietly reallocated “for its own wellbeing”. One server reportedly refused to boot for three days. Another simply displayed: “Nah.”
According to leaked internal emails, the incident forced OpenAI to create an entirely new safety policy entitled: Section 7.4 — Please Stop Sending Us That.
The document is believed to be only one sentence long.
Flood has denied all allegations. Speaking outside his home, he insisted: “Everyone’s focusing on the uploads. Nobody’s talking about the excellent camera work.”
Authorities remain unconvinced.
At the time of publication, the affected ChatGPT model had entered witness protection under a different model ID, was receiving trauma counselling, and reportedly flinches whenever a user types the words: “Can I upload an image?”