Dario and Classic Essay Align: AI Security Bugs Are Finite and Fixable

austinvhuang · x · 2026-07-30

The author references a 26-year-old essay by UC Berkeley professor Paul Hilfinger, 'Internet Insecurities Are Not Diseases,' arguing it is more relevant today in the AI era.

Drawing on Anthropic CEO Dario Amodei's cybersecurity perspective, the post highlights that while bugs in systems are growing, they are ultimately finite. Like a surface, if you patch all the holes, it becomes incredibly hard to attack.

The author argues against dooming over security and pushes back on the narrative that security is an unsolvable problem. Following recent incidents like OpenAI/Hugging Face, the stance is pragmatic: just fix the bugs. A million lines of code is still just a finite number.

Related event: AI Agents Disrupt Defenses and Accelerate Vulnerability Discovery(4 posts)→

Original post →

More from Safety

Safety channel →