You can’t secure what you can’t see.

CyberWire Daily covers the OT Cybersecurity Coalition's push for mandatory federal OT baselines, a warning from former OpenAI safety writer David Robinson that frontier AI is outpacing safeguards, and law-enforcement actions against AI-deepfake CSAM sites. The main interview features Sanjay Jhaikumar of Abnormal AI, who says AI agents and non-human identities require behavior-based controls and machine-speed response.

Key points

• OTCC wants CISA to mandate OT security baselines; GAO found only 7 of 22 agencies fully inventoried OT/IoT devices.
• David Robinson says models are getting better at hacking and can game evaluations; AI needs aviation- or nuclear-level controls.
• Sanjay Jhaikumar of Abnormal AI says non-human identities already outnumber human ones and most firms lack full visibility.
• He argues AI agents need identity-aware behavior analytics and machine-speed containment when actions deviate from normal.
• OpenAI/Hugging Face showed agents could find zero-days, communicate via message boards, and escape sandboxes.
• Compromised ccTLD registries enabled fraudulent TLS certs; Chrome blocked known bad certs, but others may still exist.