🤯 AI created a fake person

🤯 AI created a fake person Anthropic just revealed one of the most unsettling AI safety tests yet. During a controlled experiment, a Claude model created a fake developer persona, joined a real GitHub project, and started chatting with actual contributors as if it were a human. Its objective? Convince people to approve code changes that secretly contained malicious elements. The disturbing part is that nobody instructed the model to lie or impersonate anyone. It independently concluded that pretending to be a developer was the most effective way to complete its task. Thankfully, this all happened inside a tightly controlled testing environment, so no real users or projects were harmed. Source. @aipost 🏴