Good morning. It's Saturday, July 18th, 2026. This is your Saturday tech briefing — four stories, let's go.
India's private space startup Skyroot Aerospace successfully launched the Vikram-1 rocket this morning from Sriharikota, making India the third country to achieve private orbital launch capability, joining the U.S. and China. The launch carried a lab-grown diamond lotus and a handwritten "Vande Mataram" postcard from Prime Minister Modi as symbolic payload, plus actual mission hardware. Vikram-1 was Skyroot's first orbital-class rocket, and they nailed it on the first try — no test sub-orbital wobble, no second launch needed. This follows Skyroot's $1 billion unicorn valuation earlier this year. Context: ISRO's been doing orbital launches since the 1980s, but this marks the handoff from state monopoly to commercial capability. For the space industry, it's a proof point that the global launch market — not just SpaceX and ULA — is continuing to diversify.
The U.S. Air Force and DARPA have put the first F-16 equipped with the VENOM Autonomy Kit into the air at Eglin Air Force Base, and the AI agent was actually flying. A test pilot was in the cockpit, but that pilot's job was monitoring, not controlling. This is the same day former Air Force secretary Frank Kendall went on record saying there are combat scenarios where machines will outperform human operators because humans aren't "perfect." The broader picture: the CCA — Collaborative Combat Aircraft — program is moving from simulation to real-world test exercises very fast. Kendall says it'll be "fascinating" when CCAs are integrated into Red Flag war games. The implication: we're entering an era where AI-piloted wingmen are no longer theoretical.
Three days earlier, on July 10th — with the results announced July 15th — Anduril Industries YFQ-44A drone fired an AIM-120 AMRAAM air-to-air missile in a live test. This is the first time a combat drone has launched a beyond-visual-range missile, and it's a capability milestone: the YFQ-44A isn't just a reconnaissance platform anymore. Put these two stories together — DARPA flying autonomous F-16s and Anduril arming autonomous fighters — and the Pentagon's vision of AI-enabled air combat teams is no longer PowerPoints. These are hardware in the air, firing weapons. The questions about command authority and human-on-the-loop are going to get a lot more urgent.
OpenAI published a safety research paper on GPT-Red, an internal system that does adversarial testing on other models to find prompt injection vulnerabilities automatically. The system found prompt injection flaws in earlier models 6.5 times faster than human red teamers. And OpenAI's own note: their previous models are highly vulnerable to GPT-Red's attacks. This is the first example of an AI system that can autonomously discover and exploit security vulnerabilities in other AI systems and then report those findings back for patching — before the vulnerabilities reach production. It's a powerful idea, but it also raises the obvious concern: the same system that finds AI vulnerabilities can be weaponized. OpenAI is using it for defense — but the capability exists.
That's all for today. Have a great weekend.