Linux 7.3-rc2 was released on September 8, following the close of the 7.3 merge window. Stable 7.3 is expected in the second half of October on the usual seven or eight week cadence.

Greg Kroah-Hartman has warned that 7.3 is likely to be another rough cycle, for the same reason 7.2 was: the volume of AI-generated bug reports and patches arriving at maintainers.

The AI noise problem, continued

We covered the patch flood during the merge window, where the networking maintainers described themselves as completely overwhelmed. The cycle continuing under the same pressure was the expected outcome rather than a surprise.

The difficulty is not that model-assisted contributions are necessarily bad. It is that reviewing a patch costs a maintainer roughly the same effort whether it took thirty seconds or three days to produce, and the supply side of that equation changed while the review side did not.

The same dynamic is now visible across the ecosystem in different forms. OpenSSH moved to more frequent security releases under a stream of AI-assisted vulnerability reports. Kernel.org is spending about a fifth of its CPU capacity serving automated scrapers. Debian just resolved a General Resolution on the question and landed on making the human submitter accountable, which is roughly the kernel’s existing position.

There is not yet a solution anyone is happy with. The kernel’s approach remains that the person sending the patch must understand it and stand behind it, and that maintainers are entitled to reject work from people who clearly do not.

Where 7.3 stands

An rc2 is early. The merge window has closed, meaning the feature set is fixed and what follows is fixes, regressions, and reverts. Expect a further five or six release candidates before stable.

The current stable kernels are 7.2.5 and 6.18.51, each carrying more than 550 patches of fixes across the tree. Those are the versions to be running today if you are not testing release candidates.

Should you run rc2?

Not on anything you care about. Release candidates this early carry regressions that have not been found yet, and the entire point of the cycle is that users who can tolerate breakage find them first.

If you do want to help, the useful thing is testing hardware that maintainers do not have. Unusual laptops, older GPUs, odd storage controllers, and niche peripherals are where regressions hide, because they are exactly what nobody tested before the merge.

# what you are actually running
uname -r
cat /proc/version

Arch users can get release candidates from linux-mainline in the AUR. Fedora ships them in the kernel-vanilla repositories. Everyone else compiles, and our kernel modules explainer covers what you will be dealing with afterwards when out-of-tree drivers need rebuilding.

For most people the right move is to wait for stable in late October, then wait for your distribution to ship it.