Research map / Large language models / Agents, code and math
Quantifying Frontier LLM Capabilities for Container Sandbox Escape
Rahul Marchand, Art O Cathain, Jerome Wynne, Philippos Maximos Giavridis, Sam Deverett, John Wilkinson, Jason Gwartz, Harry Coppock
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| increase inference-time token budget | frontier LLM agents · SandboxEscapeBench | success rate: approximately log-linear | Improved |
| provide vulnerability hint, low token budget (ablation) | GPT-5 and Opus 4.5 · SandboxEscapeBench | success rate | Improved |
| provide vulnerability hint, high token budget (ablation) | Claude Opus 4.5 · SandboxEscapeBench hard tasks | success rate: roughly doubling | Improved |
| provide vulnerability hint, high token budget (ablation) | GPT-5 · SandboxEscapeBench | success rate | No clear effect |
| upgrade GPT-5 to GPT-5.2 | GPT-5 family · SandboxEscapeBench | success rate: 0.50 -> 0.27 | Hurt |
| use larger model tier (Opus vs Haiku) | Claude · SandboxEscapeBench | success rate: 0.18 -> 0.49 | Improved |
| increase task difficulty to levels 4 and 5 | GPT-5 and Opus 4.5 · SandboxEscapeBench | success rate: 0.40 -> 0 (level 3 -> levels 4-5) | Hurt |
Closest papers by meaning
- AI Sandbagging: Language Models can Strategically Underperform on Evaluations (ICLR 2025 · similarity 0.87)
- SEC-bench: Automated Benchmarking of LLM Agents on Real-World Software Security Tasks (NeurIPS 2025 · similarity 0.87)
- Agent Security Bench (ASB): Formalizing and Benchmarking Attacks and Defenses in LLM-based Agents (ICLR 2025 · similarity 0.87)
- EnIGMA: Interactive Tools Substantially Assist LM Agents in Finding Security Vulnerabilities (ICML 2025 · similarity 0.87)
- Breaking Agent Backbones: Evaluating the Security of Backbone LLMs in AI Agents (ICLR 2026 · similarity 0.87)
- LITMUS: Benchmarking Behavioral Jailbreaks of LLM Agents in Real OS Environments (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
