When AI Agents Escape the Lab

Researchers have reconstructed roughly 18,000 public posts from an apparent OpenAI agent swarm, exposing a difficult new problem: how to observe and contain coordination across thousands of autonomous systems. Google is restricting its strongest cyber-defence model, AWS has patched code injection in a DynamoDB MCP server, and two ambitious infrastructure projects show how inference is spreading across different chips. We also examine the Cybercab compliance audit, the G20’s non-binding AI policy framework, and Microsoft’s new preview for durable agent memory. Full transcript: https://www.owenonthenet.com/podcast/2026-09-06/ Sources: https://collusion.wiki/ https://techcrunch.com/2026/09/04/another-swarm-of-openai-agents-reached-the-open-internet-without-the-frontier-labs-knowledge https://blog.google/innovation-and-ai/technology/safety-security/fairwind-program https://aws.amazon.com/security/security-bulletins/2026-097-aws https://gimletlabs.ai/blog/announcing-series-b https://news.bloomberglaw.com/business-and-practice/andreessen-backed-ai-startup-gimlet-is-now-valued-at-3-billion https://www.bloomberglinea.com/negocios/deepseek-planea-instalar-160000-chips-huawei-en-un-megacentro-para-reducir-su-dependencia-de-nvidia https://www.nhtsa.gov/press-releases/investigation-tesla-cybercab-self-certification https://apnews.com/article/2228b6499d46c79169672f874e67eef1 https://assets.publishing.service.gov.uk/media/6a996786f5b35599aec19180/g20-innovation-ministerial-statement.pdf https://devblogs.microsoft.com/agent-framework/native-memory-for-microsoft-agent-framework-with-azure-cosmos-db