A quiet holiday weekend in the US, but not for open source: Linus Torvalds drew another line in the sand over AI-generated contributions, a critical PostgreSQL backup tool was pulled back from the brink, and over a thousand GSoC contributors started writing code. Here’s what matters.
Torvalds will be “more hardnosed” about AI-driven pull requests as Linux 7.1 RC5 ships
Linus Torvalds released Linux 7.1 RC5 today and used the occasion to warn he’s going to “start being a bit more hardnosed” about what he called pointless pull requests — many of them AI-triggered. RC5 came in larger than historical norms, filled with trivial fixes to random drivers that aren’t regressions and don’t belong this late in the cycle. This is the second consecutive week Torvalds has flagged AI as complicating kernel development: last week he declared the Linux security mailing list “almost entirely unmanageable” due to duplicate reports from researchers running the same AI tools against the same code. The broader pattern is clear: AI-assisted code review is generating real value, but it’s also flooding maintainers with noise at exactly the wrong moments. Torvalds isn’t anti-AI — he’s acknowledged LLMs are useful for testing ideas and reducing grunt work — but he’s insisting that contributors take responsibility for timing, relevance, and quality rather than fire-and-forget whatever an AI flags. For maintainers of other large projects watching this unfold, the kernel is the canary in the coal mine.
pgBackRest rescued from abandonment as AWS, Percona, Supabase, and others form funding coalition
pgBackRest, the backup and restore tool used by effectively every serious PostgreSQL deployment, narrowly avoided crisis after its sole maintainer of 13 years, David Steele, announced last month he could no longer sustain the project. The backstory: Snowflake acquired Crunchy Data (where Steele worked) last year, and Steele’s subsequent search for a role or sponsorship that would let him continue maintaining pgBackRest came up short. A coalition of companies — AWS, Percona, Supabase, pgEdge, Tiger Data, and Eon — stepped in with funding commitments, and Percona plans to bring on an additional maintainer to reduce single-point-of-failure risk. This is a textbook case of what happens when critical infrastructure depends on one person inside one company, and it’s encouraging that the ecosystem responded quickly. But “encouraging” shouldn’t obscure the uncomfortable truth: the project came within weeks of falling into unmaintained limbo, and the only reason it didn’t is that enough companies happened to depend on it enough to notice.
Google Summer of Code 2026 coding period begins — record 23,371 proposals, 1,141 contributors
The official GSoC coding period kicked off today with 1,141 contributors working across 184 mentoring organizations. The numbers this year are striking: 15,245 applicants from 131 countries submitted a record-breaking 23,371 proposals, up significantly from previous years. After 3.5 weeks of community bonding — getting familiar with codebases, aligning on standards, understanding project context — contributors are now writing code. GSoC remains one of the most effective pipelines for converting students into long-term open source contributors, and at a time when maintainer sustainability is a recurring crisis (see: pgBackRest, above), the program’s scale matters more than ever.
OpenBSD 7.9 ships — the 60th release, with up to 255 CPU cores and WiFi 6
OpenBSD released version 7.9 on May 19, marking the project’s 60th release and continuing its famously predictable six-month cadence. The headline hardware change is bumping MAXCPUS on amd64 from 64 to 255, a practical acknowledgment that server core counts have exploded. Initial 802.11ax (WiFi 6) support arrives, the SpacemiT K1 and Rockchip RK3588/RK3576 SoCs gain support on their respective architectures, and Apple Virtualization compatibility is added for arm64. On the security front, OpenSSH 10.3 and LibreSSL 4.3.0 are included, along with stricter pledge(2) and unveil(2) handling — the project’s signature security sandboxing getting even tighter. The release ships with over 13,000 pre-built packages for amd64, including GNOME 49 and KDE Plasma 6.6.4. OpenBSD’s influence has always exceeded its market share: pledge and unveil inspired hardening approaches elsewhere, OpenSSH is ubiquitous, and the project’s uncompromising stance on code correctness makes it a benchmark. Sixty releases in, the model works.
Apache Kafka 4.3.0 released with 25 KIPs and 600+ commits
Apache Kafka 4.3.0 shipped on May 22, continuing the steady cadence of the post-ZooKeeper era. The release packs 25 Kafka Improvement Proposals and over 600 commits since 4.2.0. Notable additions include KIP-1035, which adds methods to the StateStore API for managing changelog offsets (relevant for custom StateStore implementations), KIP-1250 with new metrics tracking in-memory state store key counts, and KIP-1259 introducing automatic cleanup of stale state directories on startup. Since Kafka 4.0 fully dropped ZooKeeper in favor of KRaft mode, each subsequent release has focused on operational refinements rather than architectural upheaval — exactly what you want from infrastructure software that’s quietly handling trillions of events per day across the industry. The operational improvements in 4.3 are the kind of unsexy, essential work that makes Kafka easier to run at scale.