AI News for August 12, 2026 — 7min.ai

[NEWS] Brad Lightcap leaves OpenAI after eight years to 'start something new': OpenAI's longtime COO and former CFO announced his departure Tuesday, closing an eight-year run that began in 2018.
[RESEARCH] Researchers decrypt hidden reasoning from OpenAI, Anthropic and Google APIs: Security researchers led by Alexander Panfilov recovered plaintext chain-of-thought from encrypted reasoning blocks returned by every major AI provider.
[NEWS] Gemini hits 1 billion users, Google's fastest-growing product ever: Sundar Pichai said the Gemini app crossed 1 billion monthly active users, the 14th Google product to reach that mark and the quickest to get there.
[TOOLS] SpaceXAI launches Grok Bot, persistent AI coworkers at $120/month: Grok Bot turns AI assistants into standing employees rather than on-demand chat.
[RESEARCH] Nvidia ships Nemotron 3.5 Lightning and an open router that cuts task costs to a third: Nemotron 3.5 Lightning is a 31.6B-parameter open mixture-of-experts model with 3.6B active parameters, keeping the hybrid Mamba-Transformer architecture of its predecessor.
[NEWS] Nvidia guarantees 25% of its own chips' residual value to unlock its $500B deal: To make the financing math work on the $500 billion infrastructure package it lined up with Apollo, BlackRock, Blackstone and Brookfield, Nvidia is guaranteeing up to 25% of the residual value of the chips installed in those projects.
[NEWS] Spotify will label AI artists and cut them from recommendations: An "AI Persona" badge starts appearing in mid-September on artist profiles that do not represent a real person, showing up in the profile banner, About section, Search, and track rows across playlists.
[RESEARCH] Anthropic's Riemann breakthrough ran 60 subagents through 650 ideas and 31M tokens: Anthropic detailed how its unreleased model pushed the Riemann zeta lower bound: coordinating 60 subagents that spent 31 million output tokens testing 650 different ideas over about a day and a half.
[NEWS] Zoom patches device-takeover flaw found with fewer than 20 AI prompts: Researchers at A Security uncovered a Zoom vulnerability that let an attacker run malicious code on any meeting participant's device, steal data, switch on the camera or microphone, and install malware.
[NEWS] Anthropic's Riot Platforms deal is a 20-year lease that could reach $16.1B: The $9.1 billion Anthropic-Riot deal disclosed a day earlier is a 20-year lease for 191 megawatts at Riot's Rockdale, Texas campus, structured with two extension options that could push the total to $16.1 billion.

About 7min.ai · Subscribe to newsletter