AI Finds Security Flaws But Struggles to Fix Them
Eli Ben-Sasson questioned why AI models can detect vulnerabilities in software but cannot reliably repair them, as OpenAI reported patching hundreds of Linux kernel bugs.
Updated on Nov. 18, 2026 in Cybersecurity

StarkWare CEO Eli Ben-Sasson posted on X on November 18, 2026, asking why AI models can find security vulnerabilities but cannot reliably repair them. The post came as OpenAI reported its AI pipeline discovered and patched hundreds of vulnerabilities in the Linux kernel.
Why it matters
Finding bugs rewards breadth of scanning, while fixing them requires deep understanding of how a system works, a gap that matters for software security worldwide. AI-generated patches have been observed to sometimes introduce exploitable vulnerabilities of their own.
OpenAI reported its AI pipeline discovered and patched hundreds of Linux kernel vulnerabilities, but the reported figures do not establish the share of patches that were correct. Ben-Sasson's November 18, 2026 post did not identify any specific incident, leaving the repair reliability question open.
The players
Eli Ben-Sasson
Eli Ben-Sasson is CEO of StarkWare and a co-founder of Zcash, known for his work on zero-knowledge proof systems called ZK-STARKs.
OpenAI
OpenAI is an artificial intelligence research company that developed the ChatGPT chatbot and related AI models.
The details
Ben-Sasson, who is CEO of StarkWare and a co-founder of Zcash, argued on X that AI models can flag suspicious code but struggle to repair it reliably. OpenAI's report on the Linux kernel work showed AI scanning can surface and patch flaws at scale, while AI-generated patches have been observed to sometimes introduce exploitable vulnerabilities.
Timeline
November 5, 2026: Ben-Sasson warned that AI-driven attacks could emerge ahead of quantum threats, emphasizing hash-based systems and ZK-STARKs.
November 18, 2026: Ben-Sasson posted on X questioning why AI can detect vulnerabilities but not reliably repair them.
Roadmap
Ben-Sasson's call to move toward hash-based systems and ZK-STARKs connects AI-driven threats to the proof technology StarkWare builds. Sectors that depend heavily on cryptography may see growing interest in these systems if emphasis on resilience keeps increasing.
AI tools that scan code for flaws could help developers catch bugs sooner, but unreliable patches can introduce new weaknesses. Users of open-source software such as the Linux kernel may see security updates shaped by these tools in coming releases.
The takeaway
Finding a vulnerability is only half of securing software, so AI-assisted security tools should be judged on whether their fixes hold up under review. Readers and developers should treat AI-generated patches with the same scrutiny applied to human-written code.
Further reading
Explore more in Cybersecurity for context on how AI is reshaping software defense.
Source note: This article includes information reported by Crypto Briefing.



