Daily Digest: August 13, 2026

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

Guetlein: Golden Dome in jeopardy due to 2027 funding uncertainty

Gen. Michael Guetlein said Golden Dome is in jeopardy due to 2027 funding uncertainty, warning that if lawmakers don't resolve the funding question, the program will not have money to proceed.

Golden Dome's 2027 funding is still unresolved, and Guetlein says the program has no path forward without it.

DeepSeek V4 Pro 0813

DeepSeek released V4 Pro on August 13, with API docs and benchmark coverage published on Artificial Analysis. The release drew 979 points and 422 comments on Hacker News.

DeepSeek keeps shipping frontier-class updates on roughly the same schedule as the US labs.

Grok 4.6

xAI released Grok 4.6, with benchmark analysis published by Artificial Analysis. The announcement generated 594 points and 546 comments on Hacker News.

The 546 comments against 594 points is a high ratio for Hacker News, suggesting the release drew heavy disagreement as well as praise in the comments.

Securing the AI flank to provide trusted decision-making capabilities

This sponsored piece argues that AI tools are critical components of modern warfare and must be protected to ensure they remain reliable for trusted decision-making.

It's sponsored content, so treat anything in it as marketing copy; it doesn't detail specific threats or defenses, just the general argument that AI needs protecting.

Army shifts autonomy strategy to lean more on ‘awesome vendors’

An Army official told Breaking Defense the service is shifting its autonomy strategy to lean more on vendors, since its original plan to build a software backbone in-house makes less sense now that there's real industry investment in the capability.

The Army is leaning on outside vendors because they already put real capital into the software backbone the service originally planned to build itself.

Google DeepMind: Koray Kavukcuoglu takes over in frontier AI push

Alphabet shifted day-to-day control of Google DeepMind from Demis Hassabis to deputy Koray Kavukcuoglu on August 12, 2026, following a stretch in which Gemini's flagship release slipped and Google lost ground to Anthropic and OpenAI in coding benchmarks.

Shifting day-to-day control of your AI lab to a deputy right after a stretch of losing the coding benchmark race says something about which metric leadership is being judged on.

Qwen3.8-2.4T

Qwen3.8, a 2.4 trillion parameter model with 95B active parameters in FP8, was released on Hugging Face. The release generated 673 points and 156 comments on Hacker News.

A 2.4 trillion parameter model with only 95B active is a bet that sparse MoE scaling still has headroom left.

Why patching networks against cyberattacks is ‘very scary’ in the age of AI

US officials said the long-standing practice of keeping certain operational systems continuously available without patching is becoming harder to defend as cyber threats accelerate in the age of AI.

The old defense that a system is too mission critical to patch was already weak, and AI-accelerated attacks just shortened the window that excuse depended on.

// subscribe

Get the digest in your inbox

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