On October 1, 2026, the SWC team — the fast TypeScript and JavaScript compiler written in Rust, through which Next.js, Parcel, and Deno are built — announced that it would no longer accept external pull requests, citing a flood of AI-generated code as the reason. A day later, following a comment from a GitHub employee, the project author clarified that they are exploring reopening PRs for a trusted group of contributors. This case has become one of the most vivid examples of how low-quality machine contributions are squeezing human review out of open development and forcing critical infrastructure to close its doors.

What happened

On October 1, 2026, the official SWC account @swc_rs published a short announcement on X: “Due to AI-generated pull requests, we're no longer accepting external PRs” — due to AI-generated pull requests, the project is no longer accepting external PRs. The post gathered about 375 likes and 13 replies. On October 2, the project author released a thread of four tweets: after a comment from a GitHub employee on the announcement, they reconsidered the decision and stated that they are “exploring the possibility of reopening PRs for a trusted group of contributors.” On October 3, 2026, the story reached Hacker News: the submission “SWC stops accepting external PRs due to AI-generated content” by user csmantle, at the time of processing — 1 point and 0 comments. As of now, the acceptance of external PRs in the swc-project/swc repository remains closed, and the possible reopening, according to the author's clarification, will only affect trusted contributors.

Context

To understand the scale, it is important to remember what SWC is: Speedy Web Compiler, a TypeScript and JavaScript compiler written in Rust for speed — it is built into build chains through Next.js, Parcel, and Deno, and among its users are Vercel, ByteDance, Tencent, and Shopify. The case flared up against the backdrop of a growing problem in open development: agentic tools have reduced the cost of submitting a pull request to almost zero, while review is an expensive and slow human resource, and its throughput for maintainers is finite. The flow of low-quality machine contributions has received a separate name — “AI slop,” low-quality machine-generated contributions — and it is precisely this that consumes maintainers' time on filtering out noise instead of development. The classic open-source model — open doors for any contributions, and SWC became a loud case when critical infrastructure was forced to reconsider this model.

Why this matters for the industry

For the industry, this is a rare field observation of the economics of agentic code generation: when contribution generation is almost free, and review is expensive and slow, the bottleneck becomes the signal/noise imbalance in the contribution pipeline — this is a limitation of the process and institutions, not a limit of code model capabilities. For companies dependent on SWC through Next.js and other toolchains, the signal of vulnerability: a component critical for building millions of applications is run by a small team whose process proved fragile under the pressure of machine noise. For tool developers, the case provides a validated pain point and a ready-made policy template: emergency full closure with a rollback to the “trusted contributors” model. From this, a direct demand for specific products: a GitHub App with pre-screening of incoming PRs, CONTRIBUTING policy templates against AI spam, checklists for disclosing AI use for teams. The GitHub employee's reply in the thread additionally shows that the platform is tracking the problem, and therefore native triage mechanisms are theoretically possible. If the whitelist model takes root, for new coding agents the target metric will become the “acceptability” of contributions: small diffs, linkage to an issue, explainability of the change.

Why this matters for users

Practically nothing breaks: SWC is stable, the closure of external PRs affects the inflow of changes, not working builds — projects on Next.js, Parcel, and Deno continue to build as before. If you submitted or planned a PR to swc-project/swc, you cannot contribute through the usual way now, and the possible reopening will only affect a trusted group of contributors — a random contribution will most likely pass by. The practical conclusion for a contributor from the case: describe manual work, link the PR to an issue, and honestly indicate the use of AI. If your pipeline is managed by coding agents, check if they are not sending PRs to swc-project/swc — now they will be rejected at the entrance. For maintainers of their own projects, the case gives the opposite lesson: rules against AI spam are better to write down in advance, so as not to close the repository in an emergency.

What is still unknown / limitations

The announcement has no metrics: it is unknown how many PRs were generated by AI, what share was rejected, and how much review time was spent on filtering, so the phrase “AI-generated pull requests” remains a maintainer's hypothesis, not a confirmed diagnosis. The reversal within a day after a comment from a GitHub employee indirectly indicates that the original decision was reactive, not the result of a systematic assessment. Reopening for trusted contributors is only being explored: the criteria for getting into the trusted group and the deadlines have not been announced. The assertion that other OSS projects are already adopting the whitelist model is based on a single observation, not a series of independent cases, so drawing a conclusion about a platform shift from it is premature.

Sources

Author

Look at AI, editorial team