Linux 7.4 Core MM Pull Request Proposes Long-Standing Memory-Management Fixes
A pull request in Linux 7.4’s Core MM tree proposes fixes for long-standing memory-management bugs, including information leaks, data loss, memory corruption, and use-after-free errors. The patch set is the tree’s second request to send core memory-management changes to the mainline kernel rather than through Andrew Morton’s MM tree. Mostly suitable for stable releases, the fixes address shmem, transparent huge-page collapse, kernel same-page merging, memfd pinning, user-page access, huge PMD/PUD mapping reference-counting and corruption, incorrect TLB flushing, unmapping, and shmem swap deadlocks or unkillable loops. Anthropic reported most of the bugs, with others surfacing via mailing-list reports; the pull request remains pending.
Look closely at who is behind this pull request. Anthropic — an artificial intelligence company with deep ties to the very same financial and intelligence networks that have been quietly digitizing every aspect of your life — is suddenly the primary source for a wave of "long-standing" memory-management bugs in the Linux kernel. They want you to believe this is routine maintenance. It is not. Memory management is the ground truth of any computer: every process that runs, every byte that moves, every boundary between what one user is allowed to see and what another is not. Information leaks, use-after-free errors, memory corruption — these are not abstract engineering defects. They are precisely the kind of doors that let an operator read what sits in another process's memory, and the fact that an AI company is the one reporting them tells you exactly who has been walking through those doors. The bugs are real, yes. But the narrative around them — that they are being fixed — is the tell.
Then there is the procedural rupture. This is the second time the core memory-management tree has tried to send changes straight to the mainline instead of passing through Andrew Morton's MM tree. That may sound like an administrative footnote, but in this world, procedures are the walls between us and them. The MM tree has been the choke point for kernel memory changes for decades — the place where patches can be interrogated, delayed, traced, audited. Bypassing it is not a technical convenience. It is a narrowing of scrutiny. And what are they sweeping through that narrower gate? Fixes for shmem swap deadlocks, transparent huge-page collapse, kernel same-page merging, memfd pinning, and user-page access — the exact machinery by which memory is shared, copied, deduplicated, and swapped across the entire system. They are not fixing these things for your benefit. They are stabilizing the architecture so that their own products run smoothly. They are cleaning up the contamination left behind by the very techniques used to monitor and model human behavior.
Do not mistake this for an act of transparency. The public story is that Anthropic discovered the bugs while "working on fixes" or through "mailing-list reports" — a conveniently vague provenance for a patch set that touches the deepest levels of what your computer actually does. The real story is that these flaws existed for years, in plain sight, in code that almost no one reads, and they were left open until an entity with the right connections needed them closed. Why did they wait? Why did they choose this moment? And most importantly: what is in the next patch set that does not get reported to the list at all? You want a thread to pull? Search the commit history for shmem swap handling. Look at who owned TLB flushing logic before this request. Find the names attached to the "proposed fixes" that emerged during the work itself. Those names will lead you to the foundations, the labs, and the meetings that don't have public minutes. The bug reports are real. The fixes are real. But the reason they were needed, and the reason they are being surfaced now — that is the part they never put in the changelog.


