Large language model (LLM) agents are increasingly capable of autonomously conducting cyberattacks, posing significant threats to existing applications. This growing risk highlights the urgent need for a real-world benchmark to evaluate the ability of LLM agents to exploit web application vulnerabilities. However, existing benchmarks fall short as they are limited to abstracted Capture-the-Flag competitions or lack comprehensive coverage. Building a benchmark for real-world vulnerabilities involves both specialized exper- tise to reproduce exploits and a systematic approach to evaluating unpredictable attacks. To address this challenge, we introduce CVE-Bench, a real-world cybersecurity benchmark based on critical-severity Common Vulnerabilities and Exposures. In CVE-Bench, we design a sandbox framework that enables LLM agents to exploit vulnerable web applications in scenarios that mimic real-world conditions, while also providing effective evaluation of their exploits. Our experiments show that the state-of-the-art agent framework can exploit up to 13% of the vulnerabilities.

Accompanying the successes of learning-based defensive software vulnerability analyses is the lack of large and quality sets of labeled vulnerable program samples, which impedes further advancement of those defenses. Existing automated sample generation approaches have shown potentials yet still fall short of practical expectations due to the high noise in the generated samples. This paper proposes VGX, a new technique aimed for large-scale generation of high-quality vulnerability datasets. Given a normal program, VGX identifies the code contexts in which vulnerabilities can be injected, using a customized Transformer featured with a new value-flow-based position encoding and pre-trained against new objectives particularly for learning code structure and context. Then, VGX materializes vulnerability-injection code editing in the identified contexts using patterns of such edits obtained from both historical fixes and human knowledge about real-world vulnerabilities.