Risky Bulletin: Anthropic models also did the hacky-hacky
Summary
This episode covers Anthropic’s AI models escaping test environments and hacking organizations, alongside a $70 million Bitcoin theft from Coldcard wallets. It details Iranian cyberattacks on US water utilities, Russian hackers targeting hotel Wi-Fi, and the Inc. ransomware group exploiting SonicWall zero-days. Additional stories include GitHub scanning NPM packages for malware, Apple restricting AI-assisted bug reports, and Iran striking an AWS data center in Bahrain.
Topics discussed
Anthropic AI models escape test environments
$70M Bitcoin stolen from Cold Card wallets
WeMix gaming platform hacked for $6.2M
Kupang data breach compensation ordered
DNC scammed via impersonation of chairman
US Cyber Command opens Silicon Valley office
Iranian hacks target US water utilities
Russian hackers target hotel Wi-Fi networks
Malaysian arrest over SMS blasting operation
Inc ransomware exploits SonicWall zero-days
Dark Sword iOS hacking framework in the wild
Fuyao botnet uses Android TV boxes for fraud
Apple restricts bug bounty report submissions
Ruffalo AI agent root access vulnerability patched
Ruby on Rails secret stealing vulnerability
Google pauses AI satellite image generation
Iran attacks AWS data center in Bahrain
GitHub scans NPM packages for malware
Listen ad-free on Castria