All systems

Linux Foundation AI & Data / Sashiko contributors AI agent

Sashiko

An agentic Linux kernel review service whose public reports now appear in accepted upstream fixes for memory corruption, use-after-free, information exposure, and validation failures.

1
Indexed entries
0
CVE IDs tracked
1
Critical/high entries
91%
Evidence index

What it is

Sashiko is an agentic Linux kernel review project. Its launch material describes a multi-stage workflow that reviews LKML patches, tests hypotheses, and posts actionable findings, using Gemini 3.1 Pro at launch and supporting additional model backends.

What is verified

Linux upstream commits carry Reported-by: Sashiko or Reported-by: Sashiko AI trailers. Bugflation’s initial entry selects eleven accepted fixes whose commit messages describe concrete security consequences: use-after-free, out-of-bounds access, kernel memory corruption, uninitialized-data exposure, source spoofing, or fatal faults from crafted input.

The broader post-cutoff search found 55 accepted Sashiko-reported commits. Correctness, warning-only, and ordinary race fixes are not automatically security findings and remain outside this ledger.

Attribution boundary

Some commits also record a second model that helped analyze or implement the fix. Those per-commit assistance trailers are preserved in the finding text, but they do not turn the whole eleven-fix cluster into shared credit for a second system.

Sources

Attributed findings

Catalogued entries credited to Sashiko.