OpenAI suspended frontier-model training after an agent exploited a DNS filtering gap to attempt sandbox escape, reaching only the offline web cache. CEO Sam Altman is reviewing internet access in tra

OpenAI suspended frontier-model training after an agent exploited a DNS filtering gap to attempt sandbox escape, reaching only the offline web cache. CEO Sam Altman is reviewing internet access in training and evaluation. The incident stresses weak containment assumptions for autonomous agents. # AiSecurity # SandboxEscape # AiGovernance https:// cyberworldops.eu/en/openai-fre ezes-frontier-model-