TODAY IN NUMBERS
TECHNOLOGY · SATIRE · DEPLETION FALLACY

Fixes Now Arrive Faster Than Bug Reports in One Corner of Linux, Which Dates the End of Software Flaws

Linus Torvalds says the second release candidate for Linux 7.4 is unusually large, partly because of AI fixes. One subsystem's lead developer says repairs are now outpacing reports there.

Linus Torvalds has released the second release candidate for Linux 7.4 and described it as unusually big for two reasons: some code he had left out of the first candidate because he was busy, and what he called "various AI fixes".

The first candidate was already the largest in the kernel's history, as The Register reported.

In his weekly post to the kernel mailing list, Torvalds relayed an observation from one of the subsystem maintainers. "This time around Alexei claims that on the bpf side at least the fixes are now coming in faster than the new reports," he wrote.

Alexei Starovoitov is co-creator and lead developer of eBPF, the mechanism that lets sandboxed programs run inside the Linux kernel. He was careful about how far the point extends. "I think most other areas haven't reached that point yet," he said, adding: "Don't hold your breath, but maybe cross your fingers we'll be getting there in the not entirely distant future."

The context is a year of machine-assisted bug hunting. Anthropic launched its Mythos bug-finding model in April 2026, and developers have since been turning up more software flaws than before.

The monthly totals have moved accordingly. Microsoft fixed a record 974 flaws in September 2026, its third record-setting month of the year, after 622 in July and 206 in June. Oracle has fixed 1,449 flaws in a single month.

The Linux kernel runs to more than 40 million lines of code. The Register also notes that attackers are increasingly going after flaws that AI tools have found.

The eBPF numbers invite an inference the maintainers have been careful not to draw. If fixes arrive faster than reports, the gap between them closes, and a gap closing at a steady rate closes on a calculable date.

This newsroom takes the view that a piece of software contains a fixed quantity of bugs. On that reading the figures cease to be a patching workload and become a depletion curve: Microsoft's 206, 622 and 974 are not three difficult months but three instalments of a finite debt, and the acceleration is the encouraging part.

Oracle's 1,449 in a single month puts it further ahead still. A company that fixes flaws faster than anyone else must be closer than anyone else to having none left, which makes the longest patch list in the industry the strongest available evidence of quality.

The 40 million lines are no obstacle to the argument. They are simply the size of the supply, and the largest first release candidate in kernel history indicates the rate at which it is being worked through.

Our analysts expect the crossing point to be marked by silence. When the last report has been filed and the last fix merged, the kernel will require no further releases, and a project that issues no releases is indistinguishable from one that has finished.

The Department of Unnecessary Research has adopted the lead developer's method in full. It is not holding its breath, and it has its fingers crossed, which it understands to be the project management instrument now in use.

MICROSOFT'S MONTHLY TOTALSQUESTIONABLE ANALYTICS
21June 206
64July 622
100September 974
Normalised 0-100 for display; Microsoft fixed 206 flaws in June 2026, 622 in July and 974 in September.
THE BORING FACT

In his weekly kernel mailing list post accompanying the release of Linux 7.4-rc2, Linus Torvalds relayed eBPF lead developer Alexei Starovoitov's claim that on the bpf side fixes are now coming in faster than new reports, while noting that the release was large partly because of various AI fixes.

Source: The Register
OUR PROFESSIONAL CONCLUSION

Because fixes in one subsystem now arrive faster than reports, and software is treated as holding a fixed stock of bugs, the rising patch counts become a depletion curve with a calculable end, and the longest patch list becomes the best evidence of quality.

REALITY CHECK · DEPLETION FALLACY

Unfortunately, there's a small problem.

Bug reports are a flow, not a fixed stock being drained: new code, new tools and new attention all generate new findings, and a fix-to-report ratio in one subsystem over a few weeks says nothing about the total in a 40-million-line codebase. Treating a flow as a finite stock is the trick.

THE BS DETECTOR

Where did reality end?

Tap the paragraph where this story stopped being true. Then see where everyone else bailed.

 

READER CREDULITY METER

Did you believe this story before the twist?

Be the first reader to answer.

YOU WRITE THE CONCLUSION

Think you can end this story better than our analysts?

One deadpan conclusion, 280 characters at most. Start from the facts above and take them one step too far. Our editor picks the best one each morning, and it runs in the next edition.

280 left
No real people's private lives, no tragedy, nothing that accuses anyone of anything. Our editor reads every entry and prints one; the rest are never published.