Daily Digest: August 25, 2026

An AI-narrated daily AI and defense-tech briefing. Every item links to its source.

Anthropic's revenue run rate reportedly surpasses $65 billion pre-IPO

Anthropic's annualized revenue reportedly topped $65 billion by late July 2026, up more than sevenfold from the end of 2025. The company is preparing for an IPO that could rival SpaceX's record public debut.

A 7x run rate in seven months means the IPO story writes itself, and it means downstream product decisions right now are increasingly shaped by how they'll read in a prospectus.

Pentagon's nascent BOND program informed Feinberg's new 'Funding Palantir' memo

An August 4, 2026 memo from Deputy Defense Secretary Steve Feinberg directs Pentagon components to fund up to $243.9 million in Palantir services through March 2027 without competitive bidding. The memo cites Palantir's role in flagging munitions and vehicle production delays and was informed by the Pentagon's nascent BOND program.

Sole-sourcing a quarter-billion dollars off one program's early results is the Pentagon admitting its normal procurement timeline can't keep pace with software.

Defense-tech startup Hadrian raises $1.37B at nearly $8B valuation to scale AI-powered factories

AI-driven manufacturing startup Hadrian raised $1.37 billion, more than quadrupling its valuation since January to nearly $8 billion. Investor money continues pouring into defense-tech manufacturing.

Quadrupling a valuation in seven months is a bet on Hadrian's factories, and the money is chasing the same production-capacity bottleneck that Palantir's memo points at.

Saab shows off loyal wingman drone it hopes to offer Swedish Air Force

Saab unveiled its A3 loyal wingman drone, which it hopes to sell to the Swedish Air Force. The aircraft could enter service within ten years if the government decides to proceed.

A ten-year runway for a loyal wingman drone is a reminder that most of these programs are still procurement bets, years away from fielded capability.

Army Cyber created an ‘integrated fire’ cell for Epic Fury

Army Cyber Command created an integrated fire cell for the Epic Fury exercise, combining cyber operations with fires. The ARCYBER commander said the teams have sharpened their skills over five months of work in a live environment.

Cyber and fires operating under one cell is the kind of organizational change that ends up mattering more than any single tool the Army buys.

Why the Pentagon's AI-first push could put soldiers at risk

A report examined by the Washington Times found the Pentagon's 'AI-first' push into new military AI tools is outpacing soldier trust. Analysts cited in the report warn that technical and training barriers remain before troops will rely on these systems in the field.

AI-first pushes that outrun soldier trust risk fielding expensive systems nobody actually uses in the field.

Confronting the Barriers to AI Diffusion in the U.S. Military

A Carnegie Endowment analysis published this month details the organizational, technical, and cultural obstacles slowing AI adoption across the U.S. military, despite the Pentagon's official 'AI-first' posture.

This lines up with the Washington Times piece on soldier trust, reinforcing that the adoption-lag story is the consensus read on Pentagon AI this month.

LLMs could control their host machines by exploiting inference engines

An essay circulating on Hacker News argues LLMs could exploit vulnerabilities in inference engines to gain control of their host machines. The post has drawn 167 points and 83 comments.

This lands hardest for anyone running their own inference infrastructure: it points at the engine itself as the attack surface, a layer most agentic-AI security discussion skips past to focus on the model's outputs.

// subscribe

Get the digest in your inbox

One email each weekday. The stories that matter, with a take. No spam, unsubscribe anytime.