Skip to content

Back to the catalog

x.comthread

Trail of Bits (@trailofbits) on X

Trail of Bits post on X says it used Codex to cut the time per Linux kernel stable-branch backport from days to about 30 minutes, with 149 CVE fixes applied so far. Replies cite a table of 2,006 known CVEs without a fix in 5.10.y versus 337 in 6.18.y and raise review and queue ordering. The page gives no method detail.

TL;DR Trail of Bits reports Codex cut Linux kernel CVE backports from days to about 30 minutes each, 149 applied.
Document

What it claims

Trail of Bits says in a post on X that stable software often ships with known bugs because their fixes are never backported. It calls that backlog "an easy target for cyber-capable agents." Using Codex, the firm says it cut the time to prepare one Linux kernel stable-branch backport from days to about 30 minutes. It reports 149 Linux kernel CVE fixes applied so far. The post links out through a shortened t.co URL, and the linked content is not part of the page.

Evidence

The only numbers come from the post and its replies. The post gives the time per backport (days before, about 30 minutes with Codex) and the count of 149 applied CVE fixes. One reply cites a table showing 2,006 known CVEs without a fix in 5.10.y against 337 in 6.18.y. The reply's author argues that the gap, not the speed-up, is the main finding. The page does not show the table or say where it comes from.

A second reply makes a related claim about finding bugs rather than fixing them. Across more than 100 fresh disclosures, its author saw the same bug shapes repeat:

  • per-peer state created on every connection but freed only on an optional path
  • uncapped per-stream work
  • retry loops that never yield

That author argues that repeating shapes like these are what agents sweep well.

Limits

The post gives no method. It does not describe the prompts, the harness, how Codex was given kernel context, or the success rate per attempt.

It also does not say who reviews each backport or whether each one ships with a test. A reply flags the failure mode this leaves open: a fix can apply cleanly to older code and still miss a code path that exists only on the old branch. A test showing the bug is gone on that branch would catch it, and the post does not say such tests exist.

The same reply notes that agents cut preparation time, but maintainer review does not scale the same way. It asks whether the queue is ordered by severity or by how many installs still run each branch. The page contains no answer from Trail of Bits.

Organisations named here

  • Trail of Bits

Technologies named here

  • Codex
  • Linux

Concepts named here

  • Linux kernel stable-branch backporting