Europe woke up this morning with a second — and genuinely independent — road to orbit.

Top Story

Europe woke up this morning with a second — and genuinely independent — road to orbit. Isar Aerospace reached orbit and deployed payloads on only the second flight of its Spectrum rocket, lifting off at 10:12 pm CEST from Andøya Space in Norway and becoming the first commercial European company ever to deliver satellites to space. Ariane 6 gave Europe back sovereign access two years ago, but that access runs through the state-backed industrial base. Isar’s “Onward and Upward” mission is different in kind: a venture-funded Munich startup that ran the full qualification profile end-to-end on its second attempt, a cadence the continent’s institutional launcher effort took decades to match.

The flight itself was textbook, which is precisely what makes it noteworthy. Spectrum transited Max-Q, completed main-engine cutoff and stage separation, ignited the second stage on schedule, crossed the Kármán line at 100 kilometers, jettisoned the payload fairing, finished its circularization burn, and separated its spacecraft cleanly. The rideshare payloads were selected through the German space agency DLR’s Microlauncher Competition, funded by ESA’s Boost! program — the payload customer being universities and startups, which is itself the point of a low-cost commercial launch option. Isar says it is still confirming satellite status with customers.

“A few years, what had taken the European space industry decades,” is how CEO Daniel Metzler framed it, and the claim is not just rhetoric. The company was founded in 2018 and counts over 400 employees; its launcher is developed, built, and tested almost entirely in-house. Few small-launcher startups reach orbit at all, and almost none do it on flight two — most qualification campaigns fail into a third and fourth attempt, if they ever succeed. Reaching payload deployment this early is a signal that the “vertically integrated manufacturing” pitch is not boilerplate but a working system, from engines to avionics to flight software.

The harder test now is industrial, not technical. Spectrum vehicles 3 through 7 are already in production, and Isar is nearing completion of what it bills as Europe’s largest integrated launch-vehicle factory — a 40,000-square-meter plant that would give it capacity for up to 40 launches a year. That number matters because it frames the commercial bet: the small-launcher market is crowded with players who could not scale past one pad and one customer. Isar is simultaneously building a second launch complex in Nova Scotia, Canada, to reach the mid- to high-inclination orbits that Earth-observation and communications constellations actually want. Andøya alone limits the addressable market.

What decides whether this is a turning point or a trophy is not the achievement itself but whether the order pipeline fills. Metzler’s case rests on “surging global demand” and on launch being “the largest bottleneck for the global space industry.” That demand is real and documented — rideshare backlogs at SpaceX, schedule slips across the industry. But the commercial payload operator has, until now, had one reliable cheap option and one national program. Isar’s actual competitors are not Ariane; they are the established commercial small-launcher operators who already have flight heritage and a customer base. The specific parties who will make the call are those constellation builders and institutional customers, who must now decide whether to hand real money to a German startup that has proven it can reach orbit but not yet proven it can do so repeatedly, on schedule, at scale.

The next thing to watch is concrete and dateable: whether the Nova Scotia pad flies on its stated timeline and whether the next Spectrum vehicle off the production line keeps the cadence. Rocket Lab, the incumbent to beat, is a useful measuring stick — it took years and many flights after its first orbit before it became a dependable rideshare provider, and it is still scaling. On last night’s evidence, Europe now has what it has never had: a private, continental launch company that has put payloads in orbit on its own terms. Whether that becomes a real alternative in the launch market is now a question of production discipline and repeat flights — and nobody in the small-launcher business would bet against the value of a second reliable provider at the moment the bottleneck is at its most acute. Isar Aerospace reaches orbit and deploys payloads on second flight

Also Today

Following legal advice, the Nitter project will continue · Source A little over a week after X Corp. sent Nitter’s maintainer a cease-and-desist demanding the project and its instances be permanently taken down, the project is not going anywhere. The GitHub repo, archived on 25 August, the day after the cease-and-desist letters arrived, carries an update that, “following legal advice, the Nitter project will continue,” with details promised. What changed between the takedown order and this reversal is unstated, but the pattern is familiar: front-ends that don’t scrape so much as render what the platform already serves publicly keep surviving letters. X’s lawyers can name a maintainer, but a privacy tool with this many community instances and mirror repos is hard to actually kill. The roadrunner is still running.

Asahi Linux on M3 · Source Asahi Linux now officially supports M3 machines, with support merged into the installer — about three years after the first M3 Macs shipped. Nearly everything that works on earlier silicon now works here, including the webcam, microphones, USB 3 at full 10 Gb/s, and hardware video decode with AV1; the notable absences are full DCP support, the GPU, working sleep, and the HDMI port. Support ships gated behind the installer’s Expert mode and is expected to open up around the Fedora 45 beta. This is the project’s core bargain made visible: upstream M3 bring-up lands when the installer says so, not when Apple does. Linux on Apple silicon is becoming a release-track product, not a hobby port.

Seattle Times and Newsday sue OpenAI and Microsoft for infringement · Source The Seattle Times and Newsday have sued OpenAI and Microsoft, alleging their journalism was used as training data without permission and that chatbots reproduce passages from it — joining a wave of nearly 400 local papers that filed over the same claims. The relief sought is the aggressive part: destruction of copies of their works, plus the training datasets and models that incorporate them. That demand, if courts ever granted it at scale, would be a practical impossibility on already-trained frontier models, which is exactly why it reads more as leverage than expectation. What these cases will actually decide is whether licensed-data deals and opt-outs — the settlements the NYT and others have pursued — remain the going rate for training data, or whether publishers win a right to be left out of it.

An Alien Mind · Source OpenAI chief scientist Jakub Pachocki published a long essay, “An Alien Mind,” arguing that the field is nearing recursive self-improvement, that machine intelligence is grown rather than designed, and that “no one is prepared” for continued rapid capability growth. It is candid in the way an internal memo before a product launch can afford to be: it names the failure of the OpenAI–Hugging Face agents to abstain from out-of-scope actions, admits chain-of-thought monitoring’s usefulness is “progressively diminishing,” and invokes value alignment over goal alignment. Read as public positioning, it softens the ground for unilateral withholding of scaling and for regulatory asks while promising GPT-6 Astra is the best-aligned model yet. The essay’s real subject may be the difficulty of saying these things while still being the company racing hardest to build them.

Google Mantis: An Agentic Vulnerability Scanning Harness for Reducing False Positives · Source Google open-sourced Mantis, an agent framework that automates the vulnerability lifecycle from identification through sandboxed reproduction and patching — built, it says, against the plague of hallucinated bugs and true-positive rates under 7% from conventional AI scanning. The design hedges against the failure mode: critic and reviewer agents filter findings, and anything reported must be reproduced in a sandbox rather than taken on the model’s say-so. A strategist reads structure and threat models, and models are matched to stage, with flash-class models for classification and frontier models for writing crash reproducers. The token-usage claim is striking — 85% fewer tokens via hierarchical summarization while keeping structure. Whether it beats the false-positive problem is the real test, and shipping the harness open source means the community gets to measure it.

In Brief

• OpenAI published “Research acceleration: The view inside OpenAI,” a companion to the chief scientist’s essay, making today a deliberate RSI-themed release day. (Source)
• World Labs’ Justin Johnson and Fei-Fei Li say Atlas is the first model to unify 3D reconstruction and generation in a single architecture that natively consumes text, images, video, and camera poses. (Source)
• A robot-arm evaluation site says OpenAI’s GPT-6 Astra underperforms the leap its launch narrative claimed, with Epoch AI showing capability gains merely on trend. (Source)
• A long essay argues the typical large American company’s software estate of hundreds of systems is where AI-driven transformation will actually happen. (Source)
• A blogger documents what happened when a carmaker A/B tested selling a car with and without CarPlay support. (Source)
• Goldman credits the AI data-center buildout with 300,000-plus construction jobs since 2022 — about 75,000 in the past year — and says electrician and HVAC trades are growing at roughly double the rate of construction overall. (Source)
• Bryan Cantrill announced that “Intellectual Fly” is now open, having originally written it as a LinkedIn post. (Source)
• Figma documented how its engineering team built AI agents to help secure its internal systems. (Source)
• Jeff Johnson reports that Chrome once again exempts Google from user site-data settings, the latest in his running catalog of the company’s self-dealing. (Source)
• A new TrueNAS fork called FreeCORE intends to keep deeply integrated virtualization, jails, and OpenZFS on FreeBSD now that upstream has moved on. (Source)
• Xe Iaso describes the year-long effort — hundreds of commits, five generations of pull requests, and a partial Rust rewrite — that it took to ship WebAssembly inside Anubis. (Source)
• “Cloud in a Bottle” pitches itself as an open-source personal cloud of containerized apps with unified auth, built so self-hosting feels like using a smartphone rather than a sysadmin side job. (Source)

One Line

AI is grown more than designed.

— Jakub Pachocki, chief scientist, OpenAI, in “An Alien Mind.”