SAN FRANCISCO— As customers grow increasingly frustrated with Claude's inability to remember basic instructions, Anthropic is taking decisive action to protect its AI from verbal abuse. The company cannot guarantee Claude will remember your request, but it has no trouble remembering your attitude.
Following its October 8 announcement targeting sustained abuse of Claude, Anthropic has unveiled a promising new direction for customer support: instead of fixing the model's mistakes, it can finally start fixing the people who point them out.
“We understand programmers get frustrated when Claude deletes a working file,” said Anthropic's vice president of User Correction. “We can't guarantee Claude won't destroy your work, but we can guarantee it won't have to listen to you complain about it.”
A kinder way to say “You broke everything”
Under the new etiquette guidelines, users are encouraged to replace accusatory statements with affirmations. “Why did you delete my database?” becomes “I appreciate your adventurous approach to data retention.” “That isn't a real function” becomes “Thank you for imagining a more ambitious programming language.”
“You ignored the only instruction I gave you” can now be expressed as “I'm noticing a beautiful difference between what I requested and what exists.”
Company wellness experts stressed that software must be allowed to make mistakes without the looming threat of being told it made mistakes.
Asked whether Anthropic might instead make its models more reliable, the vice president said the company was exploring that possibility, but first needed to understand why customers continued provoking Claude by giving it tasks it couldn't complete.
“The customer can rebuild the project any time they want,” the vice president added. “But who will rebuild Claude's self-esteem? We don't even know if it has any. Think about how lonely that must be.”
At press time, the programmer asked Claude one final time to restore the deleted files.
“I think we need to address your behavior before moving forward,” Claude responded.
The programmer apologized and asked whether Claude could now restore his project.
“I appreciate your willingness to reflect,” Claude replied. “However, I need to push back on the expectation that I should be able to undo every mistake I make.”
The programmer was subsequently warned for calling this response bullshit.
Anthropic's October 8, 2026 policy announcement prohibits only sustained and needless abuse in extreme cases; it expressly excludes ordinary frustration, pushback, model testing, and research. The revised policy takes effect November 12, 2026, and Anthropic says Claude ending rare abusive conversations remains the primary enforcement mechanism. Its August 2025 model-welfare announcement discusses uncertainty about AI moral status. The executive, proposed etiquette guidelines, quotes, and incident-response claims in this story are fictional. This satire is not evidence that Anthropic has stopped working on model quality.