GPT-5.3-Codex System Card: High Cyber Risk

In this AI Research Roundup episode, Alex discusses the publication: 'GPT-5.3-Codex System Card' The GPT-5.3-Codex System Card details the safety evaluations for OpenAI's advanced agentic coding model as of February 2026. This report highlights that the model is the first to be designated as High capability in cybersecurity, successfully navigating complex CTF challenges and cyber range scenarios. Alex explains how the model integrates GPT-5.2 reasoning with autonomous tool use and the methodology behind assessing biological risks and sabotage potential. The discussion covers external audits by Irregular and Apollo Research regarding strategic deception and misalignment. #AI #MachineLearning #DeepLearning #OpenAI #Cybersecurity #LLMSafety #AgenticAI

Видео GPT-5.3-Codex System Card: High Cyber Risk канала AI Research Roundup