“Sander Schulhoff predicts that within the next year, AI systems will be powerful enough to cause, and will begin to cause, tangible real-world harm due to security vulnerabilities.”