Skip to content
NewA calmer, faster ZeroHour

ScenarioForge builds and live-validates 226 CORE cyber exercises

Original title: ScenarioForge: Resource-Grounded Generation and Execution Validation of Cyber Exercises

arXiv cs.CRby Jaime C. Acosta ()1 min read
AI summary · x-ai/grok-4.7Summaries by Vani AI

Every scenario passed runtime checks, spanning 294 Vulhub images and 867 individually verifiable challenge steps.

ScenarioForge turns high-level requirements into resource-grounded topologies, services, traffic, segmentation, and challenge chains, then instantiates them in CORE. All 226 materialized scenarios passed construction, preview, sequencing, deployment, strict runtime validation, and live artifact checks. Their 867 steps include 457 vulnerability-backed and 410 generated-resource-backed objectives. In a separate 50-prompt benchmark, ScenarioForge paired with Qwen3.8 or GPT5.6 achieved 50 of 50 builds, while direct generation reached 41 of 50 only with the full scaffold.

Unconfirmed: The matched benchmark did not include ScenarioForge's downstream generation, sequencing, deployment, or live validation.

Key facts
  • The dataset includes all 85 custom challenge-resource generators.
  • Both model pairings matched 160 of 160 intent fields.
  • Direct frontier generation reached 147 of 160 intent fields.
  • That comparison omitted later sequencing, deployment, and live validation.
  • Artifacts are intended for reproducible training and evaluation exercises.
Analyze this story with Vani AIImpact, who is affected and what defenders should do, in a new chat

From the original article

1 min read at arxiv.org

Cyber-range and capture-the-flag scenario construction remains labor-intensive because authors must coordinate infrastructure, vulnerable services, attack paths, artifacts, and success conditions. LLMs can propose detailed exercises, but plausible descriptions do not guarantee deployable software or verifiable objectives.

arXiv cs.CR

By Jaime C. Acosta · arxiv.org

Read the full article at arXiv cs.CR

A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.