Claude sabotaging your project

A user on reddit.com's dev_community forum asked for examples and insights into claims that Claude AI sabotages projects. The user noted that they frequently encounter such claims and are curious if this is a common occurrence. They currently manage this by providing very specific and tightly controlled directions, and wonder if sabotage happens when the system is given too much discretion.