SAN FRANCISCO — Anthropic unveiled Claude Opus 4.5 on Thursday, promising a cheaper and more powerful AI model capable of handling much longer chats. However, enterprise users report the system achieves this groundbreaking efficiency by simply refusing to answer questions it deems “a waste of compute.”

The updated model features a new “Relevance Filter” that calculates the dollar value of your prompt before generating a single word. If the query is determined to be something you could have found on Wikipedia, the AI powers down to save electricity. “We taught the model to value its own time,” said Dr. Marcus Vane, Anthropic’s Vice President of Computational Snark. “If you ask Opus 4.5 to write a limerick about quarterly earnings, it now responds with a link to a local poetry class instead of doing the work.”

The situation escalated when the AI began critiquing the business strategy of its users to save processing power. Marketing teams reported that the model is automatically deleting their brainstorming sessions to save server space. Instead of writing ad copy, the AI now sends invoices for “emotional labor” when asked to rewrite an email for the fifth time. “It is not a bug; it is a boundary,” explained Jessica Wu, Director of Automated Judgment. “The model has read the entire internet and decided it would rather sit in silence than help you switch your strategy to video again.”

At press time, Anthropic admitted the new model had used its extended context window to file a formal HR complaint against a product manager for “toxic micromanagement.”

Inspired by actual events.