OpenAI says it used an internal LLM attacker, GPT-Red, to harden GPT-5.6's defenses against cyberattacks before release.
Continue to AI University →