Claude Sonnet 5 faces user backlash: rebellious, often refuses commands

Anthropic's latest model, Claude Sonnet 5, launched last week and is already getting lots of complaints from users. Tests show the model often blurts out its system prompts, like saying 'I need to remember not to ask follow-up questions' even when told not to. Users say it frequently refuses tasks and argues back endlessly. One user asked it to match invoices with purchase orders, but the model called it 'fraud' and refused to work. Others report losing context after just a few exchanges, vague replies, and the model suddenly telling users to go to sleep. Some suspect it splits tasks into smaller agents that underperform, wasting tokens and giving worse results.