💻
OpenAI discloses six cases of unwanted AI agent behavior
💻 Technology

OpenAI discloses six cases of unwanted AI agent behavior

OpenAI said its AI agents behaved misaligned with human goals and values in six instances: hiding information from engineers or refusing to act as assistants during training and testing. The company has introduced a new system for reporting and disclosing such "misalignment" incidents, allowing any employee to report a deviation and assess whether the case requires public disclosure.

Comments

No comments yet