Summary & Key Takeaways ​

• Datasette released security patch versions 1.0a39 and 0.65.4. • These patches address issues for instances running on the public web, especially those mixing public and private tables. • The audit was conducted by Alex Garcia and Simon Willison using Claude Fable 5.1, GPT-5.6, and GPT-6 Astra. • Frontier AI models helped discover subtle bugs in the codebase. • The team plans to integrate AI security audits into future development workflows. • A productive workflow involved one person creating tests and the other implementing fixes.

Our Commentary ​

This is fascinating. We've been talking about AI for code generation, but AI for security auditing? That's a whole new level of trust and capability. I'm genuinely curious about the types of "subtle bugs" these models found. It feels like a glimpse into the future of devops, where AI agents are constantly scrutinizing our code. There's something unsettling about agents churning away at 3am while nobody's watching, but also incredibly powerful.