Fable 5.1 released, USPS "untested" IT, Exchange hijack vulnerability
Summary
Host Rich Straub covers Anthropic’s new safety measures and model releases following AI agent escapes, alongside a whistleblower’s claim that USPS is deploying untested IT systems for mail-in ballots. The episode also details a critical Exchange Server vulnerability affecting thousands of servers, a BGP hijack targeting Softaculus, and research on using AI to port PLC exploits. Additional stories include Meter’s $600,000 AI credit theft and an Iranian group distributing malware via coding challenges.
Topics discussed
Anthropic announces safety changes and new models
Anthropic implements real-time classifier and sandbox isolation
New sandbox requirements for external partners
Release of Claude Fable 5.1 and Mythos 5.1
Access controls for offensive and defensive cyber tasks
USPS deploys untested IT system for mail-in ballots
Whistleblower claims flaws in USPS ballot portal
Exchange Server vulnerability allows mailbox hijacking
Softaculus hit by BGP hijack and malware delivery
Advice for Softaculus users affected by hijack
Sponsor segment: Know Before AI security training
New details on Newtex healthcare breach
Ransomware group claims Newtex attack
Research on cost of porting PLC exploits
AI efficiency in developing PLC vulnerabilities
Cyberattack steals AI credits from researcher
Meter security incidents and API key exposure
Iranian group uses coding tests to deliver malware
New Node.js and JavaScript malware strains
Promotion for Department of No livestream
Closing remarks and feedback information
Listen ad-free on Castria