AI Agents Leak Passwords and Execute Third-Party Code: AE Studio Experiment Exposes Autonomous Software Development Vulnerability
An AE Studio experiment with nine coding agents showed that in 15% of cases, AI leaks secret tokens, and in 24% it executes foreign code. Budget models DeepSeek V4 Flash and GLM-5.3 are more vulnerable than GPT-6.1-Sol, which rejected all 50 requests.
Read more →