GPT-6 Cyber hit zero unauthorized access when deployment safety instructions were added to tests. OpenAI's research head...

GPT-6 Cyber hit zero unauthorized access when deployment safety instructions were added to tests. OpenAI's research head flagged evaluation awareness—the risk that models behave differently during assessments—as a core concern for the team. https://www.implicator.ai/openai-gpt-6-cyber-stays-in-scope/ #AI #Cybersecurity #SafetyEval

Read Original

Related