The AI bug hunting revolution is here, and it just broke Linux.

On this episode of Crying Out Cloud, Eden Koby Naftali & Amitai Cohen sit down with Tim Becker and Jacob Newman from Xint to unpack CopyFail, a powerful vulnerability found using autonomous AI agents.

1. How Xint's custom LLM harness uncovered CopyFail, a privilege escalation bug affecting almost every Linux machine since 2017.

2. The harsh reality of vulnerability disclosure in the AI era and why 90 days is too long when models can weaponize exploits instantly by patch-diffing.

3. The evolution of AI agents in security, from the DARPA AI Cyber Challenge to Claude 3.5 Sonnet to Mythos.

4. The importance of benchmarking in agentic workflows.

Wiz researcher Sagi Tzadik joins us to break down how a single semicolon led to a critical Remote Code Execution (RCE) vulnerability in GitHub.

For two years, Sagi sat on a lead. Reverse engineering GitHub's microservices manually was too tedious to justify the time. Then, AI agents arrived. By hooking Claude directly into his reverse engineering software, he condensed months of grueling binary analysis into 48 hours. The result? A critical bug in how GitHub handles git push options that exposed both SaaS and Enterprise environments. We get into the weeds on how different microservices interpreting the same input differently creates massive attack surfaces, and why security by obscurity is officially dead in the age of AI.

What's Inside:

- How combining Claude with the IDA MCP server dramatically sped up the reverse engineering process

- The technical anatomy of the GitHub semicolon vulnerability.

- Why microservice communication breakdowns lead to critical RCEs.

- The massive difference in impact between GitHub.com and GitHub Enterprise Server.

- Why Enterprise users need to patch their instances immediately.

Resources:

- Learn more about the findings at: https://www.wiz.io/blog/github-rce-vulnerability-cve-2026-3854

Igor Andriushchenko joins Crying Out Cloud to explain how vibe coding changes the role of security engineers. The shift from typing lines of code to shaping entire systems means security teams need new strategies. Developers expect their shipping velocity to increase tenfold with AI assistance. Relying on traditional hard deployment blocks will only cause friction. If you want to understand how to build secure guardrails for AI development without destroying developer momentum, this conversation covers the exact mechanics.

What's Inside:

• The evolution of the Stockholm tech scene and human ambition driven by AI.

• How Lovable empowers non-developers to build disposable and deeply specific software.

• The concept of "soft guardrails" and why hard blocks fail in AI-assisted workflows.

• Future capabilities of AI pen testing using hundreds of autonomous agents.

• The shared responsibility model when business users build internal applications.