Skip to content
Source · Daily Brief

AI Morning Brief — 12 August 2026

The consumer AI race hit a symbolic parity point overnight as Google confirmed Gemini has passed one billion monthly active users — its fastest-growing product ever — matching ChatGPT, which crossed the same threshold in June. Separately, security researchers published findings showing that the encrypted reasoning traces returned by all major frontier model APIs can be extracted and replayed, with public sessions already containing leaked passwords and API keys.

Top stories

  • Gemini and ChatGPT both at one billion users. Google CEO Sundar Pichai announced the milestone, calling Gemini the fastest product in Google history to reach one billion monthly actives. OpenAI’s ChatGPT had crossed the same threshold in June, making this the first time two AI assistants simultaneously stand at that scale. via The Verge
  • Stolen reasoning traces from frontier APIs. Researchers found that Anthropic, OpenAI, and Google return encrypted chain-of-thought blocks to clients that can be replayed across sessions, users, and different models. A scan of public sessions surfaced dozens of passwords and API keys; the traces also show that reasoning summaries visible to users often conceal what the model is actually doing. via Simon Willison
  • Anthropic commits to watermarking all Claude outputs. Claude-generated text will carry embedded watermarks and generated files will include C2PA provenance metadata, partly to comply with European AI transparency rules. The policy applies globally, covers new models from August 2026 onward, and Anthropic plans to release detection tools for third-party verification. via The Verge
  • Brad Lightcap leaves OpenAI after eight years. Lightcap, who served as COO before moving to a special projects role, posted an internal memo to X saying he would be starting “something new.” His exit follows the recent departure of OpenAI’s head of ethics, who left last month without a replacement being named. via The Verge
  • OpenAI begins testing ads in ChatGPT. The company announced a limited ad test to help sustain free access, with promises of clear labeling, answer independence, and user controls. The move arrives as OpenAI also wrapped a $7 billion employee stock buyback at an $852 billion valuation ahead of a widely expected IPO. via OpenAI
  • River AI raises $1.1 billion at two months old. The personal-agent startup, founded by xAI co-founder Igor Babuschkin, secured a General Catalyst-led round at an age unusual for any sector. The company is focused on long-horizon personal agents. via TechCrunch

Who shipped

NVIDIA released Nemotron 3.5 Lightning, a 30B latent mixture-of-experts model with 3.6 billion active parameters and up to 670 tokens per second, framed explicitly for always-on agentic workloads, alongside NeMo Switchyard, an open-source LLM router for distributing agent tasks across models. Meta made Muse Glimmer 30B available as open weights — a dense multimodal reasoning model that fits on a single consumer GPU, described by observers as the first meaningful open-weight release from Meta since the original Llama era. OpenAI was unusually active: the ads test launched alongside Daybreak cybersecurity models on Amazon Bedrock and a native ChatGPT desktop app for Linux. Microsoft shipped MAI-Code-1.1-Flash in GitHub Copilot with added vision support, while deprecating MAI-Code-1-Flash effective 10 September.

Open-source pulse

Muse Glimmer 30B from Meta drew immediate attention in local-inference communities, with early benchmarks showing competitive coding performance against Qwen 3.6 27B at similar memory requirements. Nemotron 3.5 Lightning shipped in NVFP4 and BF16 variants with a 1M context window and native tool use. LTX-2.5 launched as an open-weights video world model targeting local NVIDIA hardware with multishot scene coherence. The community is also anticipating Qwen 3.8-27B, confirmed by the Qwen team for this week.

Money, infra & hardware

Anthropic signed a $9.1 billion data center lease with Bitcoin miner Riot Platforms in Texas covering 191 megawatts, with extension options that could push the total to $16.1 billion. via THE DECODER. NVIDIA announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to mobilize over $500 billion in AI infrastructure capital, guaranteeing up to 25 percent of its own hardware’s residual value to win over investors. via THE DECODER. Anthropic’s planned IPO — at a reported $965 billion valuation — is facing investor questions about Chinese competition and political headwinds ahead of a September or October filing.

Quiet corners

DeepSeek posted nothing official overnight; the community is benchmarking V4 Flash 0731 and speculating about a forthcoming V4 Pro release, but no announcements came from the lab. Cohere was absent entirely. Mistral published only a European infrastructure note — regional inference capacity and new compute, no model updates.

By the numbers

  • 305 stories in 24 h across 40+ sources
  • Most-mentioned model: GPT
  • Most-mentioned lab: OpenAI and Anthropic
  • Notable absences: DeepSeek, Cohere

Compiled by AI Feed’s editor from all 305 headlines published on 11–12 August 2026.