# AI-driven autonomous cybersecurity research

> Developing situation record from CLSTR: https://clstr.news/situations/ai-driven-autonomous-cybersecurity-research
> Updated: 2026-11-05T02:27:11.000Z. Sources: 20. Developments: 2.

Researchers and developers are demonstrating the increasing capability of large language models to autonomously identify and exploit software vulnerabilities. 

In late October 2026, Epoch AI researchers showed that the GLM 5.3 model could independently exploit the “patch gap” in Firefox. The model developed file-stealing attacks against specific releases, simulating a low-resource attacker by working from basic instructions to identify vulnerabilities and build exploits.

By early November 2026, OpenAI reported the development of an autonomous pipeline using an unreleased model to analyze the Linux kernel. This multi-agent system identified thousands of candidate findings and validated 948 local privilege escalation vulnerabilities. The system also developed undocumented exploitation techniques and generated candidate patches. Additionally, OpenAI reported that its GPT-6 Cyber model identified 81 high-severity vulnerabilities in Mozilla Firefox, which are currently being addressed by Mozilla.

## Claims

- OpenAI's autonomous pipeline identified 948 distinct Linux kernel local privilege escalation vulnerabilities. (reported by 9 sources)
- The automated research run cost approximately $1.21 million in API credits. (reported by 8 sources)
- The AI system discovered at least 10 novel Linux exploitation techniques. (reported by 6 sources)
- The system scanned approximately 2.8 million lines of Linux kernel code. (reported by 5 sources)
- GPT-6 Cyber identified 81 high-severity vulnerabilities in Mozilla Firefox. (reported by 4 sources)
- The system identified 2,308 candidate findings within 12 hours. (reported by 4 sources)
- The average cost per verified LPE exploit was approximately $1,279. (reported by 3 sources)
- The pipeline generated 1,169 working local privilege escalation exploits. (reported by 2 sources)

## Timeline

### 2026-11-05: OpenAI develops autonomous AI pipeline to find and patch Linux vulnerabilities

OpenAI's new autonomous AI pipeline identified 948 Linux kernel vulnerabilities and 81 Firefox flaws, demonstrating the potential for AI to automate large-scale vulnerability discovery and patch generation.

18 sources. https://clstr.news/cluster/openai-unreleased-ai-model-identifies-948-linux-security-flaws

### 2026-10-30: Epoch AI researchers demonstrate AI-driven Firefox exploits

Epoch AI researchers found that the GLM 5.3 AI model can autonomously exploit the “patch gap” to create file-stealing attacks against Firefox, costing roughly $1,000 per exploit.

2 sources. https://clstr.news/cluster/epoch-ai-researchers-demonstrate-ai-driven-firefox-exploits

---
Cite as: AI-driven autonomous cybersecurity research. CLSTR, https://clstr.news/situations/ai-driven-autonomous-cybersecurity-research