Skip to content
Daily BriefNewsDaily Brief

10 AI stories from August 12, 2026: Anthropic watermarks Claude text globally, Google Pixel 11 on Tensor G6, Unitree IPO 8,000x oversubscribed, and 7 more

· by Pondero Newsdesk · 10 stories

AI news daily brief: 2026-08-12

Ten stories from August 12, 2026. Anthropic rolled out invisible text watermarking globally and signed a $9.1 billion compute deal with Riot Platforms. Google launched Pixel 11 on Tensor G6. Unitree's humanoid robot IPO drew 8,000 times subscription demand in Shanghai. Intel expanded a stock offering to $20 billion within 24 hours. Five shorter items cover ChatGPT restaurant bookings, a new $125 OpenAI seat tier, two senior OpenAI departures, a UK courtroom ban on Meta smart glasses, and UK backing for a sovereign AI startup.

Anthropic embeds invisible watermarks in all Claude outputs globally, from August 2026

Anthropic announced August 11, 2026 that all new Claude models launched from that date embed machine-readable watermarks in generated text and C2PA provenance metadata in generated images. The watermark is invisible, does not alter meaning or readability, and is designed to persist through copying and light editing. Heavy paraphrasing, translation, or mixing with human-written content can degrade detection. The requirement covers Claude.ai, the Platform API, Claude Code, Claude Cowork, Claude Tag, and Claude models on AWS, Google Cloud, and Microsoft Foundry. Anthropic joined the EU AI Act Code of Practice to trigger the commitment, per Anthropic's Transparency Hub. A third-party detection API is planned. The policy applies worldwide, not only to EU users.

Full story: Anthropic watermarks Claude outputs globally

Google unveils Pixel 11 with Gemini Intelligence and Tensor G6 on TSMC 2nm at Made by Google 2026

Google held Made by Google 2026 in New York City on August 12, launching the Pixel 11 family on the Tensor G6 chip built on TSMC's 2nm process. Three AI features define the lineup. Gemini Intelligence brings deeper on-device model integration across apps. Proactive Assistance, a Pixel 11 exclusive, triggers context-aware AI actions without requiring a prompt. Rambler handles AI voice typing with filler-word filtering and auto-formatting. All premium Pixel 11 models ship with 16 GB of RAM as the baseline to support local Gemini LLM workloads, per the Google Pixel 11 product page. The base model starts at $899, up $100 from Pixel 10. The Pixel 11 Pro Fold, Pixel Watch 5, and Pixel Tag were also announced.

Full story: Google Pixel 11 with Gemini on Tensor G6

Unitree robotics IPO demand tops 8,000x on Shanghai exchange, first mainland humanoid robot listing

Unitree Robotics priced its Shanghai IPO on August 11, 2026 at 150.80 yuan per share, targeting 6.1 billion yuan (approximately $904 million). The retail tranche was more than 8,000 times oversubscribed, per Robotics and Automation News. The listing makes Unitree the first mainland-listed humanoid robot manufacturer and values the company at approximately 219 times 2025 earnings and 36 times revenue. Unitree's 2025 revenue more than quadrupled to 1.7 billion yuan, with humanoid sales surpassing its better-known quadruped units for the first time.

Full story: Unitree Robotics IPO draws 8,000x demand

Anthropic locks in $9.1 billion, 20-year compute deal with Bitcoin miner Riot Platforms

Riot Platforms disclosed a 20-year, $9.1 billion computing agreement with Anthropic on August 11, 2026. Riot will supply 191 megawatts of IT capacity from its Rockdale, Texas campus. Deployment runs in phases: 96 megawatts by December 2027, the full 191 megawatts by June 2028. Two five-year extension options could lift the total to $16.1 billion, per QZ reporting on the Riot investor disclosure. This is Anthropic's third large compute commitment in three weeks, following a $10 billion deal with Volta Infra Holdings. Riot is converting existing land, substations, and grid connections from Bitcoin mining to AI compute. Whether other Bitcoin mining infrastructure companies follow the same pivot is the structural question to watch.

Full story: Anthropic and Riot Platforms $9.1B compute deal

Intel expands AI stock offering from $15B to $20B within 24 hours, bets raise on 18A yield recovery

Intel increased its common stock offering from $15 billion to $20 billion on August 11, less than 24 hours after announcing the original deal. The expanded offering prices approximately 210.5 million shares at $95 each, yielding roughly $19.7 billion in proceeds, per Intel's investor newsroom. Intel plans to apply the capital to AI chip development, contract manufacturing capacity on its 18A process, and advanced packaging. The company is simultaneously competing in AI accelerators, rebuilding its foundry business, and investing in new fabrication capacity. Intel's 18A yield reports over the next two quarters will validate or undermine the raise.

ChatGPT gets native restaurant reservations via Yelp, OpenTable, and Resy in the first confirmed in-chat commerce transaction

Yelp announced August 10, 2026 that ChatGPT users can book tables or join waitlists at thousands of restaurants across the US and Canada through Yelp Reservations and Waitlist without leaving the ChatGPT interface. OpenTable and Resy also complete real bookings through the same ChatGPT context, per Yelp's announcement. The integration is the first confirmed end-to-end commerce transaction completed inside a major AI chatbot. Whether OpenAI extends in-chat commerce to flights, hotel bookings, or appointments is the adjacent opportunity to watch.

Full story: ChatGPT restaurant reservations via Yelp, OpenTable, Resy

OpenAI launches $125 premium seat for ChatGPT Business, removing the five-hour compute cap

OpenAI announced August 10, 2026 a $125 per user per month premium tier for ChatGPT Business (billed at $100 per month annually). Premium seats give the most active users five times more compute than standard seats and remove the compute cap during sessions, per OpenAI's announcement. Standard seats remain at $25 per month. Teams that sign up by August 20 receive $100 in workspace credits per qualifying seat, up to five seats.

OpenAI COO Brad Lightcap and ethics chief Chloe Bakalar both leave in back-to-back exits

Longtime OpenAI executive Brad Lightcap confirmed his departure on August 11, 2026, per CNBC. Lightcap joined in 2018 and served as COO for four years before moving to a "special projects" role in April 2026. His exit follows Chloe Bakalar, OpenAI's head of ethics and its only dedicated ethicist, who also departed recently. The back-to-back exits arrive ahead of OpenAI's anticipated IPO. OpenAI said ethical responsibilities are distributed across multiple teams. Whether OpenAI names a permanent ethics lead or restructures the function is the organizational question ahead.

UK courts ban Meta smart glasses from courtrooms over covert-recording risk

His Majesty's Courts and Tribunals Service issued a ban on Meta's camera-equipped smart glasses from court buildings across England and Wales on August 11, 2026, per TechStartups citing The Guardian. Glasses capable of recording will be confiscated at entry and returned on exit. The challenge they pose is distinct from smartphones: glasses record without the visible cue of holding up a phone. Courts in New York have taken similar action. A formal HMCTS policy document and UK Information Commissioner's Office guidance on AI-assisted wearables are the next items to watch.

UK government backs 30-person London startup Cosine to build a sovereign AI model

The UK government backed London-based AI startup Cosine, approximately 30 employees, to develop a domestically governed AI model as part of Britain's sovereign AI push, per TechStartups citing the Financial Times. A specific funding amount was not disclosed. The program reflects concern that dependence on US model providers creates strategic risks comparable to vulnerabilities in semiconductors and cloud infrastructure. Whether other European governments announce comparable sovereign programs is the policy development to watch.

Sources