gpt-5.1

Researchers introduced institutional red-teaming, a methodology for testing deployment rules in multi-agent AI systems, and found that changing only the consequence rule shifts mean fatality by 22 to …