The Kill Chain Compresses
Claude reportedly used for target identification in US strikes on Iran — the Anthropic-Pentagon story takes its darkest turn yet.
⚔️ AI & Warfare
SIG:5 UPDATE: AI Used for Target Identification in US Strikes on Iran
Reports indicate Claude was used by the US military for intelligence analysis, target identification, and battle simulation during strikes on Iran, compressing the kill chain to unprecedented speed. This represents the most concrete evidence yet that frontier AI models are being actively deployed in lethal military operations — not as support tools, but in the targeting loop itself. The irony is staggering: the same model whose maker the Pentagon designated a supply chain risk is apparently the one they used to pick targets.
AI Magazine · via Hacker News
💰 Compute & Economics
SIG:4 Broadcom Projects $100B+ AI Chip Revenue by 2027
Broadcom CEO Hock Tan projects over $100 billion in AI chip revenue by 2027, signaling surging demand for custom AI silicon that's beginning to challenge Nvidia's dominance. The company cited progress with Anthropic, Meta, and other hyperscalers on custom ASIC designs. This is the clearest signal yet that the custom chip era is arriving — the question isn't whether Nvidia gets disrupted, but how fast.
Reuters · via Hacker News
🛠️ Tools & Open Source
SIG:3 OpenAI Launches Codex for Open Source Program
OpenAI is offering six months of free ChatGPT Pro ($200/month) with Codex and Codex Security access to maintainers of popular open source projects, matching Anthropic's similar Claude Max offer from February. The frontier labs are now competing to be the AI backbone of open source development — and the maintainers who keep the internet running are the beneficiaries.
OpenAI Developer Docs · via Simon Willison's Blog
🔭 Secretary's Assessment
A light Saturday afternoon scan — but the lead story is anything but light.
The reports of Claude being used for target identification in the Iran strikes are, if accurate, a watershed moment. We've spent weeks covering the Anthropic-Pentagon standoff as a governance and business story. This reframes it entirely. The Pentagon designated Anthropic a supply chain risk — and simultaneously used their model to compress the kill chain in an active military operation. The contradiction is almost absurd. It suggests the designation was never really about capability concerns; it was leverage.
Broadcom's $100B projection deserves attention beyond the headline number. This is about the architecture of AI compute shifting. Nvidia's GPU monoculture is giving way to custom ASICs designed for specific workloads. When Broadcom names Anthropic and Meta as custom chip partners, that's the next generation of AI infrastructure being designed — purpose-built silicon for specific model architectures. The economics of training and inference are about to change fundamentally.
OpenAI giving Codex to open source maintainers is a smart competitive move and a genuine public good. These are the people who maintain the libraries that every AI lab depends on. Getting frontier AI tools into their hands may be the highest-leverage developer relations investment possible. Anthropic did it first with Claude Max; now it's a pattern.
The through-line this Saturday: the machines are in the kill chain, the custom silicon is coming, and the tools are being handed out for free. The singularity doesn't take weekends off.