How Anthropic’s silence fueled a Claude Code trust crisis

Anthropic silently degraded Claude Code for seven weeks through three engineering missteps: reducing reasoning effort from high to medium on March 4, a bug discarding reasoning history mid-session on March 26, and a system prompt capping responses at 25 words on April 16. Users noticed declining performance — including more errors, instruction-following failures, and riskier 'edit-first' behavior — but received no communication until a postmortem appeared after the final fix on April 20. Engineering leaders and developers argue that silence amplified speculation and eroded trust, and that real-time status updates during incidents are more valuable than detailed postmortems published weeks later.