What changed
OpenAI announced the release of ChatGPT Cyber, a model specifically designed for cybersecurity applications, which will be accessible to the highest tier of users in its Daybreak Red program. This program provides offensive cybersecurity researchers with tools for vulnerability hunting and development. The model has already been used to scan and prepare patches for vulnerabilities in the Linux kernel, identifying 948 potential issues, and found 81 'novel' vulnerabilities in Firefox, the web browser.
What this means in practice
The new model is intended to expand Daybreak's capabilities, enabling organizations of all sizes, independent researchers, and open-source maintainers to engage in advanced security research. OpenAI emphasized its goal of dramatically expanding Daybreak to address the growing threat landscape. However, the model's availability is limited to the highest tier of users, suggesting a controlled rollout to mitigate potential misuse.
Limitations and availability
OpenAI has also noted that the Daybreak Blue program, which previously provided access to powerful frontier models, has been 'nerfed' to make it less effective at finding vulnerabilities. This change may be an effort to limit access to potential bad actors, though cybersecurity professionals argue that such restrictions could hinder their ability to use frontier technology while malicious actors can deploy open-weight models with fewer guardrails. The company has stated that only 'thousands' of cybersecurity professionals are participating in the Daybreak program, compared to a much larger number of potential hackers.
