Anthropic Got Hacked Again: How Ex-DOGE Researchers Got Into Its Slack With AI Agents
Summary
This episode analyzes a security breach at Anthropic, where a one-month-old startup called Cathedral exploited a third-party Salesforce integration to access sensitive Slack data. The host discusses the political tensions between Anthropic and the Pentagon, noting that Cathedral’s co-founder is a former Pentagon data head who recently secured a government contract. The discussion highlights how AI agents can now perform continuous, automated attacks using older models, exposing the fragility of current software security and the urgent need for perpetual, AI-driven defense systems.
Topics discussed
Introduction: Anthropic breach and the rise of AI hacking
The Cathedral startup and the bug bounty program
Exploiting Salesforce to access Anthropic's Slack
Political tensions: Pentagon, Anthropic, and blacklisting
Government response and the national security angle
Expert criticism: Storing critical data in Slack
Anthropic's defense and the need for annual security reinvention
The challenge of rebuilding global software security infrastructure
The 24/7 AI defense model and the 'Clash of Clans' analogy
Implications for consumers and small businesses
The danger of third-party supply chain vulnerabilities
The 24/7 nature of AI attacks vs. human limitations
Government affiliation and the path to nationalization
Conclusion: Low-cost attacks and global security risks
Outro and call to action
Listen ad-free on Castria