I’m legitimately considering cancelling my Claude subscription. Over the past few months, I’ve noticed that Anthropic’s Claude, specifically its Sonnet 5.5 and Opus 5.5 models, has become increasingly restrictive when it comes to the work I want to accomplish. What frustrates me isn’t AI safety as a general concept, nor am I making a criticism of AI assistants across the board. My issue is specifically with Anthropic’s implementation of restrictions in Claude, particularly when those restrictions interfere with otherwise legitimate productivity and development tasks.

To give you two examples. First, document processing. I have a PDF copy of a book that I managed to obtain online. The document contains considerable amounts of unnecessary white space, and I wanted to have it reformatted and recompiled into a Word document to make it more practical to read and work with. When I attempted this using Claude, it refused on the grounds that the material was copyrighted. However, I was able to accomplish precisely what I wanted using both ChatGPT and Grok. Neither presented the same obstacle. This raises an obvious question: why is Claude refusing a document-processing task that competing AI assistants are perfectly capable of handling?