The systems that no one will test
a day ago
- Author found a vulnerability in a Brazilian federal system in 2020, gaining access to personal data of over 200 million people, and reported it responsibly.
- By 2026, rapid ML advances and scaling of RL environments raise concerns about AI agents replicating similar flaws in unprotected government systems.
- The vulnerability required only attention to detail, not extraordinary expertise, making AI agents potentially more dangerous as they are faster and scalable.
- ML is a planetary problem requiring global thinking beyond technophobia/technophilia, with emerging economies most vulnerable to such AI-driven cyber threats.