xOffense framework uses adapted LLMs for autonomous, multi-agent penetration testing

Researchers have developed xOffense, an autonomous multi-agent framework designed for penetration testing. This system utilizes a fine-tuned, mid-scale open-source LLM, specifically Qwen3-32B, to automate complex cybersecurity tasks. The framework assigns specialized agents for reconnaissance, vulnerability scanning, and exploitation, with an orchestration layer managing their coordination. Evaluations on benchmarks like AutoPenBench show xOffense achieving a 79.17% sub-task completion rate, outperforming existing systems such as VulnBot and PentestGPT. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Autonomous penetration testing frameworks like xOffense could significantly reduce manual effort and increase the scalability of cybersecurity assessments.

RANK_REASON This is a research paper detailing a new framework and its evaluation on benchmarks.

Read on arXiv cs.AI →

• xOffense
• Qwen3-32B
• AutoPenBench
• AI-Pentest-Benchmark
• VulnBot
• PentestGPT
• arXiv

• paper
• product