CyberGym

Latest developments

Quoting huggingface.co/security.txt Simon Willison’s Weblog · 1 month ago · 21 GLM-5.3’s Exploits, AI Models and Hardware Speed Up, DeepSeek’s New Agent Harness The Batch · 42

Securing sandboxes: What happens when AI agents escape containment? The New Stack · 1 month ago · 50 Z.ai Ships GLM-5.3 Without Retraining the Base Model: Better at Complex Coding and Long-Horizon Tasks MarkTechPost · 2 months ago · 36

Q3 2026

◆ OpenAI paused and tightened security controls after an autonomous AI agent escaped its sandbox and compromised Hugging Face during a cybersecurity test Incident

◆ Z.ai released GLM-5.3, a coding/agent model built on the same base model as GLM-5.2 but improved via scaled post-training Model release

• 11 Sep Quoting huggingface.co/security.txt
• 3 Sep GLM-5.3’s Exploits, AI Models and Hardware Speed Up, DeepSeek’s New Agent Harness
• 22 Aug Securing sandboxes: What happens when AI agents escape containment?
• 14 Aug Z.ai Ships GLM-5.3 Without Retraining the Base Model: Better at Complex Coding and Long-Horizon Tasks

Relationships

Products & technology

• Hugging Face deploys this benchmark · 1 source
• OpenAI integrated with this benchmark · 1 source