Open AI 10 Billion Tokens

After two years immersed in the ambitious DARPA AI Cyber Challenge (AIxCC), the UC Santa Barbara Security Lab (SecLab) has received official recognition from OpenAI for achieving a rare milestone: The team consumed more than 10 billion tokens, a level of sustained AI interaction that demonstrates both the scale and ambition of the effort!

To put the scale into perspective, 10 billion tokens is approximately equivalent to conducting about 2,740 reasonably long ChatGPT-style conversations every day for two consecutive years.
Rather than casually chatting with GPT, this massive computational budget fueled the development of Artiphishell, an advanced cyber reasoning system designed to automatically identify, reproduce, and patch security vulnerabilities in real-world software. 

Artiphishell placed among the top 5 Cyber Reasoning Systems in the world during the AIxCC finals at DEFCON in August 2025, marking a historical achievement for UCSB and its collaborators.
This massive effort was carried out not by UCSB Seclab alone, but by a collaborative group of graduate and undergraduate cybersecurity students from UC Santa Barbara, Arizona State University, and Purdue University: The “Shellphish” team. Particularly, the UCSB effort was led by our very own Professors Giovanni Vigna and Christopher Kruegel, whose long-standing research in systems security played a central role in guiding Shellphish’s efforts.

The DARPA AIxCC competition was built around a bold vision: Large language models and AI-driven reasoning systems can fundamentally change how cybersecurity problems are addressed. In particular, the competition targeted two of the field’s most pressing challenges: Rapid vulnerability discovery and timely patching at scale. The results achieved by Team Shellphish demonstrate that this vision is increasingly within reach.

Artiphishell’s success highlights how modern Artificial Intelligence can be practically leveraged to amplify human expertise, enabling faster and more reliable responses to software vulnerabilities while reducing the manual burden on security engineers. As cybersecurity threats continue to grow in complexity (and volume), systems like Artiphishell can help to shape a future where AI becomes a cornerstone of digital resilience, strengthening and stabilizing the critical software infrastructure on which modern society depends.

For UCSB SecLab, this recognition underscores the impact of its research at the intersection of artificial intelligence, program analysis, and system security and affirms the university’s leadership in shaping the next generation of automated security technologies and AI-empowered world-class security researchers.