Public articles linked to the same research event.
arXiv The work proposes CVE2AP, which combines structured prompting with planner-driven error feedback so an LLM can automatically generate PDDL-encoded attack paths from natural-language CVE descriptions; evaluated on 21 CVEs, five models and 12 configurations, the best model GPT-5.5 reaches 86.9% syntax correctness, 78.6% solvability, 53.8% embedding similarity and 93.1% LLM-as-expert semantic correctness, with error feedback giving the most consistent gains and GPT-5.5 the best quality-cost trade-off.
The work proposes CVE2AP, which combines structured prompting with planner-driven error feedback so an LLM can automatically generate PDDL-encoded attack paths from natural-language CVE descriptions; evaluated on 21 CVEs, five models and 12 configurations, the best model GPT-5.5 reaches 86.9% syntax correctness, 78.6% solvability, 53.8% embedding similarity and 93.1% LLM-as-expert semantic correctness, with error feedback giving the most consistent gains and GPT-5.5 the best quality-cost trade-off.
The work proposes CVE2AP, which combines structured prompting with planner-driven error feedback so an LLM can automatically generate PDDL-encoded attack paths from natural-language CVE descriptions; evaluated on 21 CVEs, five models and 12 configurations, the best model GPT-5.5 reaches 86.9% syntax correctness, 78.6% solvability, 53.8% embedding similarity and 93.1% LLM-as-expert semantic correctness, with error feedback giving the most consistent gains and GPT-5.5 the best quality-cost trade-off.
The work proposes CVE2AP, which combines structured prompting with planner-driven error feedback so an LLM can automatically generate PDDL-encoded attack paths from natural-language CVE descriptions; evaluated on 21 CVEs, five models and 12 configurations, the best model GPT-5.5 reaches 86.9% syntax correctness, 78.6% solvability, 53.8% embedding similarity and 93.1% LLM-as-expert semantic correctness, with error feedback giving the most consistent gains and GPT-5.5 the best quality-cost trade-off.