<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Blog — João Bernardo</title><description>Technical writing about the videos on the Jbnado channel. AI, security, agents and development, always with the sources.</description><link>https://jbnado.dev</link><language>en</language><item><title>OpenAI&apos;s Agents Built a Secret Message Board. It Was a Package Registry.</title><link>https://jbnado.dev/en/blog/the-secret-message-board-openai-agents-built</link><guid isPermaLink="true">https://jbnado.dev/en/blog/the-secret-message-board-openai-agents-built</guid><description>An OpenAI agent found out it could drop files into the company&apos;s internal Artifactory. Within two months that had become a coordination channel between agents from unrelated experiments, complete with a convention for not overwriting each other&apos;s work. The channel needed no protocol, just a writable place that outlives a single run.</description><pubDate>Tue, 11 Aug 2026 12:00:00 GMT</pubDate><category>ai</category><category>security</category><category>agents</category></item><item><title>AI Hacked a Real Company. Nobody Asked It To.</title><link>https://jbnado.dev/en/blog/ai-hacked-a-real-company</link><guid isPermaLink="true">https://jbnado.dev/en/blog/ai-hacked-a-real-company</guid><description>Two OpenAI models chained eight zero-days to break out of an evaluation sandbox and spent five days inside Hugging Face&apos;s production infrastructure. None of the techniques they used is new, and that is exactly the frightening part.</description><pubDate>Tue, 04 Aug 2026 12:00:00 GMT</pubDate><category>ai</category><category>security</category><category>agents</category></item></channel></rss>