AI Briefing

Today's throughline is a tug-of-war over who controls AI and how open it should be. Nvidia's Jensen Huang broke his X silence to champion open-weight models, while Anthropic quietly wired another $20 million into a group lobbying for guardrails, and Google shipped a cyber-defense model it's keeping locked to governments only. Meanwhile Substack is betting readers will pay a premium just to know a human actually wrote the thing.

01ProductEngadget ↗

Substack Now Scans Your Posts to Tell Readers How Much Was Written by AI

Substack has rolled out an AI-detection feature, built with startup Pangram, that scans any post, note, or comment over 100 words and shows an estimated split between human-written, AI-assisted, and AI-generated text. Creators also get a new space to disclose how they use AI. It matters because Substack is effectively betting its business on the idea that readers will pay more for writing they know a human produced, at a moment when AI-generated content is flooding every feed. Critics warn the detector could false-flag genuine writers, and it can't tell when AI was used purely as a research aid.

02IndustryFortune ↗

Jensen Huang Breaks His X Silence to Defend Open-Weight AI, and OpenAI and Anthropic Sit It Out

Nvidia CEO Jensen Huang made his first-ever post on X to share an open letter, signed by 25 companies including Microsoft, Meta, and Palantir, defending open-weight AI models as essential to American safety, innovation, and 'sovereignty.' He drew a direct parallel to the 1980s fight over open-source software, warning Washington not to over-restrict the technology. The subtext is a live policy battle over whether to clamp down on powerful Chinese open models like Moonshot's Kimi K3. Tellingly, OpenAI and Anthropic, who see strong open-source rivals as a competitive threat, declined to sign.

03PolicyAnthropic ↗

Anthropic Doubles Down, Sending Another $20 Million to a Group Lobbying for AI Rules

Anthropic is donating another $20 million to Public First Action, a nonpartisan group that pushes for AI safeguards, doubling its total support to $40 million. The company says the money is strictly for public education and policy work and cannot be used to back or oppose any candidate for office. It matters because it's one of the clearest signs yet of a frontier lab spending real money to shape regulation ahead of the 2026 elections. Anthropic is explicitly arguing that transparency alone is no longer enough given how fast model capabilities are advancing.

04ProductThe Hacker News ↗

Google Ships an AI That Hunts Software Bugs, and Keeps It Locked to Governments

Google DeepMind released Gemini 3.5 Flash Cyber, a specialized model that discovers, validates, and patches software vulnerabilities, and is restricting it to governments and trusted partners via its CodeMender agent in a limited-access pilot. In testing on the V8 JavaScript engine, it found 55 confirmed vulnerabilities, beating both standard Gemini 3.5 Flash (47) and Anthropic's Claude Opus 4.6 (36), including 10 issues no other model caught. It matters because the frontier labs are now openly racing to build AI that defends critical code, a direct answer to fears that the same capabilities could just as easily supercharge attackers.