OpenAI Launches ChatGPT Cyber for Automated Codebase Vulnerability Hunting

01AI-Driven Auditing at Operating System Scale

OpenAI has made ChatGPT Cyber available to top-tier participants in its Daybreak Red cybersecurity program. The announcement arrives alongside disclosures that frontier reasoning models are already surfacing critical vulnerabilities in foundational open-source codebases.

Using GPT 5.6 Sol via Daybreak Blue, independent security researcher Max Justicz identified an unpatched vulnerability in Ubuntu/Debian server packages within two hours by prompting the model with a previously resolved vulnerability and asking for equivalent or worse failure modes. OpenAI reported that ChatGPT Cyber was similarly deployed to audit the Linux kernel, identifying 948 potential issues, alongside 81 novel vulnerabilities in Mozilla Firefox.

02Economics of Automated Security Research

Automated vulnerability hunting at this depth demands substantial compute. OpenAI reported spending approximately $1.2 million in compute tokens on the Linux kernel bug hunt and announced a $1 billion initiative to support defenders securing critical infrastructure such as electrical grids and municipal water utilities.

However, researchers actively participating in Daybreak report friction around model guardrails. While frontier labs seek to prevent malicious exploitation, defensive researchers note that aggressive filtering limits defensive discovery while malicious actors experiment with abliterated open-weight alternatives.