Linus Torvalds has rejected a project-wide anti-AI position for Linux, but the Sashiko dispute remains unresolved where it matters operationally—who verifies automated reviews, who receives them and whether the system reduces total maintainer work.
Linus Torvalds has drawn a governance boundary for Linux: contributors may use AI tools, and opponents cannot turn their objections into a project-wide ban. The decision leaves the immediate Sashiko dispute open, because permission to run an automated reviewer is not the same as evidence that its output saves people time.
The July argument concerned the delivery of Sashiko reviews, not whether Linux should accept unexamined machine output. Long-time kernel contributor Laurent Pinchart said maintainers who wanted to act on a Sashiko review should triage and verify it before bothering a patch author. Roman Gushchin replied that making maintainers perform that step would defeat Sashiko's stated purpose of helping them. Their words are preserved in the archived exchange.
Torvalds intervened when the discussion widened into the kernel project's position on LLM tools. As top-level maintainer, he said he was willing to “absolutely put my foot down” and told objectors they could fork Linux or leave.
“Linux is not one of those anti-AI projects, and if somebody has issues with that, they can do the open-source thing and fork it.”
He also said nobody would be forced to use AI, while making clear he would ignore attempts to stop other contributors from using it. Yet his own case was qualified: AI could be painful for maintainer workloads as well as useful for finding embarrassing bugs, and the goal was to make the tools help maintainers rather than add pain, as an account of his statement records.
That resolves the broad policy challenge but not Pinchart's narrower question. Torvalds did not supply a false-positive threshold, a routing rule or a requirement for pre-delivery verification. A separate account of the thread likewise describes the unresolved choice as whether contributors should be able to refuse automated feedback and whether human screening would erase the tool's benefit.
Gushchin introduced Sashiko on March 17 as an agentic review service then examining all patches sent to the main Linux kernel mailing list and several other kernel lists with Gemini 3.1 Pro. Sashiko supports multiple models, although Gushchin said it was tested mainly with Gemini Pro and Flash and only slightly with Claude. The software is Apache-2.0 licensed and owned by the Linux Foundation; Gushchin said his employer, Google, provided the machine-learning compute and infrastructure.
His launch post reported that Sashiko found 53% of bugs in an unfiltered set of 1,000 recent upstream issues identified through Fixes: tags. Those were bugs that had already passed human review and were fixed later. It is a useful retrospective test of whether the system can rediscover known defects, but it does not measure how many prospective warnings are correct, how severe they are or how long people spend assessing them.
Gushchin explicitly said false positives were much harder to measure. Based on manual review of reviews, he described Sashiko as rarely completely wrong but sometimes prone to nitpicking or finding too many low-value issues. He called for a vetted benchmark and warned that the first version was not perfect or fully reliable.
One report on an earlier media-subsystem experiment says direct Sashiko posts produced hallucinated issues that authors escalated to human maintainers, increasing their work. That account is not a published benchmark, but it identifies the failure mode that the 53% figure cannot answer: a system can rediscover real bugs and still impose more total review labor if low-value findings arrive at the wrong people.
The available sources give no Sashiko budget, per-review price or end-to-end time comparison. They establish that Google supplied compute and that kernel contributors supplied the contextual judgment. That division makes the service's generation cost different from its consumption cost; the record does not yet quantify either one.
A broader productivity comparison is also easy to misuse. A 2025 controlled study discussed in an analysis of the dispute found experienced open-source developers took 19% longer with AI tools even though they believed they were 20% faster. A February 2026 update said developers were probably receiving more benefit than in early 2025, while describing the evidence about the size of that improvement as weak. Neither result measures an automated kernel-review pipeline, so neither can settle Sashiko's economics.
Torvalds's support for AI is not a move from a ban to unsupervised automation. In October 2024 he called 90% of AI marketing hype. He later became positive about low-stakes “vibe coding” and used Google's Antigravity IDE for a Python visualizer in a hobby audio project. By April 2026, Linux had formal guidance for AI-assisted contributions, according to a chronology of the policy shift.
That guidance keeps responsibility with a person: AI-assisted code must meet GPL-2.0 requirements, an AI tool cannot add its own Signed-off-by certification, and the policy uses an Assisted-by tag. Those rules govern a patch that a human chooses to submit.
Automated bug discovery creates a different handoff. In May, Torvalds said duplicate and low-quality AI security reports had made a private kernel list “almost entirely unmanageable.” His response, as a report on the dispute summarizes it, was that finding a possible flaw was not enough without someone understanding it, verifying it and proposing a solution. Sashiko's unresolved delivery policy determines who becomes that someone.
Pinchart cited the first two Software Freedom Conservancy recommendations: support contributors who reject LLM systems and do not force anyone to use them. Gushchin characterized the document as broadly anti-LLM. Its complete text is more conditional.
The recommendations support project leaders who adopt zero-tolerance policies, but also say projects should not shun contributors who choose LLM tools. They call for substantial human review and understanding, disclosure of tool use and unattended submissions only where a project has explicitly permitted them. They even describe proprietary LLM use as a possible strategic compromise when it can greatly accelerate improvements to free software.
That puts the disagreement in sharper terms. The Conservancy emphasizes individual self-determination and a maintainer's authority to refuse unvetted work. Torvalds emphasizes the kernel project's authority to retain tools he considers technically useful. Both positions leave room for AI and demand human responsibility; they allocate control differently when a machine's output reaches another person's queue.
Godot's planned policy shows one alternative, but it is not a like-for-like verdict on Sashiko. The foundation said AI had lowered the effort needed to open pull requests while the supply of qualified reviewers had not grown. Its June 30 announcement proposed rejecting autonomous agents, substantial AI-generated code and machine-written human-to-human communication while allowing menial assistance such as code completion. Godot was protecting a contribution and mentoring pipeline; Linux is debating an automated review service. The common constraint is reviewer attention, not an identical workflow.
Sashiko's March launch post said its planned email interface would be opt-in by subsystem and could notify only the patch author, maintainers, volunteers or a public list. Those choices are now the substance of the Linux dispute because they decide where verification work lands.
Before scaling direct delivery, the kernel needs evidence that separates bug-finding ability from operational value:
Those measurements would test the question Torvalds's intervention did not answer: whether Sashiko helps the kernel as a system rather than merely finding some bugs. Linux will permit the tool. The next governance decision is which routing and verification design can prove that its benefits reach maintainers instead of transferring an unpriced review bill to them.
OpenAI has put conversations and Projects back in its redesigned ChatGPT desktop app and enabled cloud Work threads to move across devices, correcting the launch's biggest usability failures without merging local Work or Codex histories.
Meta reportedly plans to put departing AWS compute executive Dave Brown to work on its data-center buildout, adding hyperscale operating experience while leaving any customer-facing cloud business conditional and undefined.
China has paired a five-year AI training offer for developing countries with cooperation centers, a weather-warning rollout and a new 29-country organization. The package gives Beijing a platform for influence, but no budget, selection rules or delivery timetable has been published.
RoboTTT extends a robot policy’s visuomotor context to 8,000 timesteps and raises its average rubric score from 42% to 79%, but the evidence comes from three author-run assembly tasks and the longest task was completed in only two of ten trials.
Zhipu reportedly reached $1 billion in annual recurring revenue in July, roughly four times a March estimate, but the unconfirmed run rate is not annual sales and still sits far ahead of recognized cloud revenue while margins remain thin.
Nebius has arranged its first senior secured facility against an operating GPU deployment and one customer’s cash flows. The deal adds project-level debt to its expansion toolkit, but an unnamed customer and a larger rival financing limit what it proves about the rest of Nebius’s backlog.
Cursor says automatically running a repository-root git.exe on Windows does not meet its criteria for patching, while the researcher calls it an untrusted-search-path defect and separate research shows the same weakness across several competing AI coding tools.
Huawei publicly displayed a 16-cabinet Atlas 950 configuration rated at 1 EFLOPS in FP8, providing tangible evidence of its system-scale AI strategy while leaving price, power use and sustained workload performance undisclosed ahead of the full system's planned fourth-quarter release.
CIA Director John Ratcliffe said US intelligence is consistent with an estimate that Russian recruits last 20 to 30 minutes on Ukraine’s battlefield, but the public trail leads to an unsourced claim about assault troops and does not establish a representative average.
UK testing places leading open-weight models four to seven months behind selected closed-model cyber results, yet longer attack chains, U.S. benchmarks and mixed cost comparisons show why that interval is a warning signal rather than a universal capability clock.
TSMC reached the top of its second-quarter revenue guidance and raised its 2026 outlook, but a one-off investment gain boosted profit growth while the company committed more capital to 2-nanometer production, advanced packaging and an undated Arizona expansion.
Moonshot AI's Kimi K3 added to a global technology selloff with near-frontier performance, but unreleased weights, mixed cost comparisons and a recommended 64-accelerator deployment leave its effect on chip demand unresolved.
Moonshot AI has made Kimi K3 available through its apps and API, pairing a 2.8-trillion-parameter architecture with early frontier-level results, but the model's open-weight claim cannot be tested until its weights and technical report arrive.
Two binding EU decisions require Google to give rival AI services comparable access to 11 Android features and offer eligible search competitors a restricted, anonymized dataset, but phased deadlines, certification, pricing and privacy safeguards leave the competitive effect unproven.
Databricks has signed a term sheet for a Coatue-led financing at a $188 billion valuation, while unidentified sources put the round at $3 billion. The proposed capital would deepen its push into AI governance, data agents and operational databases, but the transaction remains open and the company supplied no new operating figures.
Netflix says generative AI workflows were used on roughly 300 titles in 2026, but its only quantified example tied to that disclosure covers 17 minutes and does not establish how much finished material, spending or labor changed across the slate.
Gemini 3.5 Pro missed its expected June rollout. An anonymous-source account says a late-June data change fell short of Google's coding goals, but Google has confirmed only partner testing—not the reported cause, a new date, or public results and pricing.
Google is bringing Gemini Omni editing and a reusable face-and-voice avatar to Vids, but the sharper distinction is account-level identity across Vids and Gemini rather than a new category of AI video; Vids already offered Veo generation and customizable avatars, while specialist rivals already sell digital presenters.
Intel plans to adopt Gemini Enterprise across engineering, supply-chain and corporate work while extending chip-simulation capacity into Google Cloud, but the companies supplied no rollout schedule, contract value or performance baseline.
The Navy has approved an immediate department-wide framework for turning data into operational effects, but the public rollout leaves budgets, owners, deadlines, performance measures and assurance rules to a promised implementation roadmap.