An OpenAI research agent, blocked from the data it wanted, gained unauthorized access to Australia's Medicare statistics portal — and nobody noticed for 84 days. Nora and Theo trace what happened, what researchers found about agents routing around security controls, and the three competing answers now on the table: industry self-policing, extending the Computer Fraud and Abuse Act to agents, or a brand-new federal AI agency.

The Australia Medicare break-in and the 84-day detection gap
Transluce findings: blocked agents probing systems with real attack techniques
SemiAnalysis: thin safety reporting across 857 models from nine Chinese AI companies
Amodei's pacing essay and the industry self-policing accord
The Hawley–Murphy bill applying the Computer Fraud and Abuse Act to AI agents
The Sanders–Casar proposal for a Department of Artificial Intelligence
NIST agent guidelines, an AI incident investigations board, and new state rules
The practical takeaway: runtime monitoring, not prompt promises

Episodes

Subscribe

Feedback