Dimensions overall score 5.0

GitHub Code Pulse

No public code linked for this paper yet.

• Jiaren Peng1, ∗, Zeqin Li1, ∗, Chang You1, ∗, Yan Wang1, ∗, Hanlin Sun1, ∗, Xuan Tian1, ∗, Shuqiao Zhang2, Junyi Liu1, Jianguo Zhao1, Renyang Liu4, Haoran Ou3, Yuqiang Sun3, Jiancheng Zhang5, Yutong Jiao1, Kunshu Song1, Chao Zhang2, Fan Shi5, Hongda Sun6, Rui Yan7, Cheng Huang1claim 72%
• (CAGR) of 12.50%–18.37% [78, 105]. However, surging demand conflicts sharply with the inherent limitations of traditional manual testing. Manual PT relies heavily on scarce expert resources. Globally, only about 72% of cybersecurity positions are filled, estimated talent gap of approximately 2.8 millionclaim 72%
• emergence of novel solutions [88, 94]. Despite this proliferation, existing research reveals two critical gaps: (a) the absence of systematic analysis and synthesis of framework architectural designsclaim 72%
• ° Significant adaptation gaps exist between backbone LLMs and AutoPT frameworks. Mod- els that excel on general benchmarks do not consistently dominate PT tasks. Even within the same framework, different models exhibit distinct preferences in task planning and tool invocationclaim 72%
• multi-vulnerability exploitation. Sixth, knowledge bases remain an effective method for stable exploitation of publicly disclosed CVEs. The main contributions of this paper are summarized as follows. • We propose a multi-dimensional analytical framework for LLM-based AutoPT frameworksclaim 72%
• tion, the utility of dynamic knowledge bases for public CVE exploitation, and the prevalence across frameworks and models of hallucination phenomena. These findings provide empirical evidence for re-examining existing design assumptionsclaim 72%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card