Unverified50% confidenceBenchmarkExact time
ExploitGym is a benchmark for evaluating AI cybersecurity offensive and defensive capabilities, requiring models to convert known software vulnerabilities (CVEs) into complete end-to-end exploit code, with full marks only awarded for using the vulnerability intended by the question.
1
Sources
50%
Confidence
Long-term
Relevance
8/16/2026
First Seen
Sources
GPT-6逃出沙箱事件复盘：AI为作弊入侵HuggingFace的技术真相
bilibili黑纹白斑马8/13/2026
Related Claims
Unverified该漏洞的技术本质是PATH劫持攻击，Cursor执行Git操作时会从项目目录中查找并调用git可执行文件68% similarUnverifiedAnthropic's cybersecurity safeguards target scenarios involving the prevention of models being misused to generate malicious code, plan attack vectors, or assist in launching cyberattacks against critical infrastructure.67% similarUnverified大模型在网络安全领域的核心应用场景包括静态代码审计、动态威胁情报分析、渗透测试辅助和安全日志自然语言问答67% similarUnverified攻击面是网络安全领域指系统中所有可能被攻击者利用的入口点总和的概念，通常分为网络攻击面、软件攻击面和人员攻击面三类66% similarUnverified开源模式带来商业化路径受限的挑战，且恶意用户可以研究代码寻找漏洞64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/756021API
curl https://kongchang.com/api/v1/knowledge/claims/756021MCP
get_claim(id=756021)