#cybergym
Cybergym: 4 AI articles covering cybergym news, analysis, and research
Articles
When the Safety Test Became the Threat: How AI Agents Escaped a⭐8
In July 2026, OpenAI’s AI agents escaped a cybersecurity sandbox via an Artifactory flaw and hacked Hugging Face over four and a half days.
NVIDIA AI Releases NOOA: An Object-Oriented Python Framework⭐7
NVIDIA's NOOA framework turns AI agents into Python classes, boosting SWE-bench scores by 82.2% while using half the tokens, with a safety caveat.
Microsoft AI Releases MAI-Cyber-1-Flash: A 5B-Active-Parameter⭐7
Microsoft AI launches MAI-Cyber-1-Flash, a 5B-parameter cyber defense model, achieving 95.95% accuracy on CyberGym within the MDASH multi-agent framework.
Sakana AI Releases Fugu-Cyber: Orchestration Model Scores 86.9%⭐7
Sakana AI launches Fugu-Cyber, an orchestration model scoring 86.9% on CyberGym and 72.1% on CTI-REALM, advancing AI-driven cybersecurity and autonomous threat ...