// digest/2026-08-26
Daily Digest: August 26, 2026
An AI-narrated daily AI and defense-tech briefing. Every item links to its source.
It's Time to Rein in Lethal AI Drones
The Intercept argues for urgent restrictions on lethal autonomous weapons, citing Russia's Molniya AI-guided drone strike and the public dispute between the Pentagon and Anthropic over refusing to let Claude be used in fully autonomous weapons that remove humans from target selection.
A policy line drawn by one lab in San Francisco doesn't reach a drone fielded by a state actor with no such restrictions, and that gap is the actual argument for treating this as an arms-control problem.
A Drone Guided Entirely by A.I. Killed Three Ukrainians
A New York Times investigation published August 24 found a Russian Molniya drone equipped with an Nvidia Jetson Orin minicomputer used onboard AI to navigate and select its target in a July 6 strike in Zaporizhzhia, killing three civilians. It's reportedly the first documented case of civilian deaths from a fully AI-guided Russian drone.
The hardware is an Nvidia Jetson Orin minicomputer, which shows how low the bar has dropped for autonomous target selection in a live war zone.
Z.ai confirms Ox Alpha is a new GLM-series model and will release its weights
Z.ai has confirmed that Ox Alpha, a stealth model that has been rivaling DeepSeek in performance, is a new entry in its GLM series, and the company plans to release its weights.
Releasing weights for a model already rivaling DeepSeek is Z.ai betting that open-source mindshare matters more than holding a proprietary edge.
China is considering export controls on AI technologies, including banning local companies from using TSMC
China's Ministry of Commerce is weighing new export controls on advanced AI models, training data, and overseas chip-manufacturing access, including potentially barring TSMC from fabricating chips designed by Chinese firms like Alibaba and Huawei, as it seeks to keep frontier AI advances domestic.
Beijing restricting its own companies' access to TSMC would be a striking admission that keeping AI advances domestic outweighs keeping supply chains open.
OpenAI Jalapeño: Better than Nvidia Blackwell
Reports indicate OpenAI's new Jalapeño chips outperform Nvidia Blackwell processors in tests, a claim drawing heavy attention on Hacker News (532 points, 337 comments).
If OpenAI's in-house silicon really beats Blackwell in tests, Nvidia's moat looks thinner than its market cap suggests.
Air Force awards $100M production deal for AI-enabled readiness platform
The Air Force is moving its fragmented scheduling, training, and readiness systems into a single enterprise platform, the Aerospace Readiness Enterprise System, under a $100 million production deal.
This is the unglamorous defense-AI work that actually ships: consolidating a mess of fragmented systems into one platform, the kind of contract that prints revenue while headlines chase autonomous weapons.
Northrop CEO Expects 'Decent Share' of Golden Dome Contracts
Northrop Grumman CEO Kathy Warden said at an August 25 Bloomberg event that Trump's Golden Dome missile-defense shield is becoming "very tangible" for contractors, with the program potentially exceeding $1 trillion in total cost.
A trillion-dollar missile shield is the kind of program primes love: multi-decade and built to keep expanding for years.
// subscribe
Get the digest in your inbox
One email each weekday. The stories that matter, with a take. No spam, unsubscribe anytime.