The brief

Wednesday, August 12, 2026

Claude, GPT, and Gemini have a critical flaw: encrypted reasoning traces extracted via weaker sister models expose in plaintext. Meanwhile, Zoom's annotation feature has a critical RCE exploitable in under 20 AI prompts—patch immediately.

12 stories · 3 podcasts · get this by email each weekday

Today's stories

I finally understand why handing an AI agent your Chrome session is terrifying
A Zoom Screen-Sharing Bug Let Anyone Take Over Other Devices on a Call
New surveillance tech links your phone to your license plate
Zoom flaw let an attacker take over your device, including iPhone and Mac
North Korean remote IT staffer worked for US government agency, says FBI
‘Zoomsday’ hack uncovered using fewer than 20 AI prompts
Woman Pulled over at Gunpoint Twice After Flock Camera Glitch
FBI says cybercriminals are hacking into victims’ online accounts to steal their intimate pictures
Researchers find that feeding a frontier model's encrypted reasoning traces to a weaker model from the same provider can make it output the traces in plaintext (Will Knight/Wired)
DEF CON crowd suspected in fake-hotspot attack on Delta flight
Researchers say suspected Chinese hackers used open-source AI agents to build an autonomous hacking tool that compromised Taiwanese government websites in July (Tom Wilson/Financial Times)
Agentic security: Enterprises enforce agent permissions two-thirds of the time — and isolate high-risk agents less than one in five

Worth a listen

Shifts We Are Seeing Across Social Engineering, Post-Disruption Impact Report
Are We Seeing The End Of OSS Models?
Measuring Sustainability via Project Kepler, with Niki Manoledaki

← All editions