The new model can find and exploit software vulnerabilities on its own, and has proven adept at evading its creators' safety monitors.
Inside the growing panic over recursive self-improvement, and why leading scientists warn we are gambling with our lives.
