The systems that no one will test
Summary
Christian Perone reflects on a real vulnerability he discovered in a Brazilian federal system in 2020 and uses that incident to discuss broader risks as AI/ML systems scale. He argues that rapid advances in reinforcement learning environments and self-improvement loops amplify security and governance challenges for nation-states, noting OpenAI’s incident reports and the idea that many systems may never be AI-pentested. The piece advocates planetary thinking in AI governance and urges safeguards to prevent systemic weaknesses from being exploited at scale.