Greg Kroah-Hartman – Security in the LLM Age [video]
Linux kernel developer Greg Kroah-Hartman debunked claims that Anthropic's Mythos model found 79 serious vulnerabilities, showing most were invalid, already fixed, or trivial.
summary is enough The talk's core debunking of AI security claims is fully captured in the summary and comments, saving you a watch.
The article could not be read automatically; this analysis is based on the title and discussion.
summary
Kernel developer Greg Kroah-Hartman dissected Anthropic's widely publicized claim that its Mythos model discovered 79 vulnerabilities in the Linux kernel during a talk. GKH showed that the majority of these reports were non-bugs, made up data, or already patched, leaving only about 10 minor issues that took a single hour to fix. Commenters echoed his skepticism, noting that current LLM security findings rely on basic pattern-matching of historical code rather than novel security research.
key points
- Out of 79 reported Mythos vulnerabilities, GKH showed 24 had no details, 14 were not bugs, 3 were fabricated, and 15 were already fixed.
- The actual valid bug fixes required only one hour of kernel development time, highlighting severe marketing exaggeration.
- Commenters observed that LLM vulnerability tools largely regurgitate historical patches via pattern matching rather than performing original analysis.
who should read it
Systems engineers and security professionals tracking the intersection of artificial intelligence and software vulnerability research.
discussion skeptical
consensus
Commenters widely agree that AI capability announcements suffer from extreme marketing hype, contrasting corporate press releases with GKH's grounded engineering reality.
pushback
A few commenters argue that rapid pattern-matching agents still pose novel automation threats and will inevitably improve over time.
Greatly appreciated the candor. I've included a few slides into text that i thought were eye-opening to me:
From his Kernel Recipes 2026 slide on Mythos
```
Mythos's 79 vulnerabilities: 24 - no detail at all "something crashed" 14 - not a bug at all 3 - totally made up data 15 - already fixed in latest release - 11 by others - 4 by anthropic 20 - fixes were needed - 7 "assume a malicious filesystem image" - 2 "assume you can inject a malicious network packet into the middle of the stack" - 2 "NOMMU" - 6 sctp networking issues for untrusted devices - 2 ipv6 minor network issues - 1 gpu …
It transcribes the exact breakdown slide from GKH's presentation, showing how the 79 claimed vulnerabilities vaporize under developer scrutiny.
Generated by an LLM from the article and thread. Check the sources before relying on it.