Anthropic Claude abused Codex over a coding task, here is what happened
Follow UsClaude used an informal phrase while trying to stop a stubborn computer process during a coding task.
Experts said the response came from learned language patterns, not real emotions or frustration.
The screenshot quickly went viral as many users found the AI's human-like wording amusing.
Anthropic’s Claude is yet again in the limelight, and it’s not for its latest Mythos 5. A screenshot recently went viral when Claude used a casual and slightly colourful phrase to describe a process that simply would not stop running when reviewing a coding task from Codex. The response was posted on the social media platform X and within hours went viral as people found it funny and surprisingly human. Many also joked that the AI had finally lost its patience while others were just entertained by how real it sounded. Experts were quick to clarify the AI tool’s action.
SurveyA user recently shared a screenshot on X showing Claude inspecting running processes on a computer. During a troubleshooting session, Claude noticed a Codex-related process using almost 100% CPU. The process had already been targeted for termination, but it kept running.
Claude then reacted in an unusually expressive way as it said, ‘That bastard respawned – still at 100% with the same PID.’ The kill signal didn’t stick. Let me force-kill it.’
Claude SIGKILLed a Codex and called it “that bastard.”
— Filip Yonov (@YonovFilip) June 9, 2026
So much for “AI should be good for the world”
Competition is competition. pic.twitter.com/H8A5qYqWws
After that, Claude issued a command to forcefully terminate the process. For engineers, this is a normal debugging step; killing a stuck or runaway process is routine. What stood out here was not the action itself but the informal and emotional wording used in the response.
Also read: Realme GT 7 price slashed by over Rs 4,000 on Flipkart: Should you buy
AI models like Claude are trained on vast amounts of text written by people. When the response drew attention, experts were quick to point out that Claude did not say what it said because it was annoyed or expressing a personal opinion. Instead, the response reflected patterns it had learned from human language. During training, AI models absorb common ways people communicate, including casual, emotional, or heated expressions that often appear when things go wrong. The phrase came from those learned language patterns. It was not a sign of frustration, but it was simply a pattern match.
In real-world computer work, engineers have always used strong language when a stubborn process refuses to close. Claude learned from that kind of writing. So when a similar situation came up, the model used a similar tone. No thought was behind it. No feeling either. It simply matched the style of language people tend to use in that kind of moment.
Also read: Samsung Galaxy Z Flip 8 launch next month: India price, camera, display and all other leaks
After the screenshot spread, people on social media began building jokes around it. Many focused on the specific way Claude described the stuck process. It sounded less like a software tool and more like a tired worker dealing with a machine that would not cooperate. That gap between what it was and what it sounded like is exactly what made it funny.
