AI BRIEFING · 2026-08-19

OpenAI seeks to one-up Anthropic with new customer privacy protections

plus 9 more stories

Also: Unitree Robotics Surges in Shanghai Debut, and Google packs Search and Gemini with new AI study tools. Every story links to its primary source.

Curated by Gaurav · August 19, 2026

01

OpenAI seeks to one-up Anthropic with new customer privacy protections

TechCrunch AI ↗

OpenAI is previewing Private Safety Processing, a service for select enterprise customers that monitors for abuse across multiple sessions while retaining zero customer data. Instead of storing conversations, the system sends OpenAI a narrowly defined signal about the type of activity detected, and staff decide whether enforcement is warranted before anyone accesses customer data. The move is a direct answer to Anthropic's July policy of holding user session data for 30 days on covered models, which had unsettled enterprises working with sensitive information.

02

Unitree Robotics Surges in Shanghai Debut

Bloomberg Technology ↗

Unitree Robotics, formally Yushu Technology, soared as much as 629 percent on its Shanghai trading debut, rising from an IPO price of 150.8 yuan to as high as 1,100 yuan after raising 6.1 billion yuan, about $904 million. It is the first humanoid robot maker listed in mainland China, and the offering was oversubscribed more than 8,000 times, a record for the STAR market. The Hangzhou company shipped over 5,500 humanoid robots last year and grew revenue to 1.7 billion yuan in 2025 from 393 million yuan, closing with a market value near 342 billion yuan.

03

Google packs Search and Gemini with new AI study tools

TechCrunch AI ↗

Google rolled out a set of AI study features across Search and Gemini, including interactive 3D simulations for topics like acidity scales and DNA structures, custom practice quizzes in science, math, humanities and languages, and Lens support that lets students photograph a problem and get a worked explanation of where they went wrong. Gemini gains a dedicated student hub that gathers flashcards, quizzes and study notebooks in one place, plus multi-step research reports that run in the background and can be discussed by voice. Google is pitching the package against ChatGPT and study apps such as Knowt and Gauth, with features rolling out in phases over the coming weeks.

04

AI Chip Startup Fractile in Talks for $6.5 Billion Value After Anthropic Deal

Bloomberg Technology ↗

British inference chip startup Fractile is in advanced talks to raise about $600 million at a $6.5 billion pre-money valuation, more than six times the roughly $1 billion it was worth after a $220 million round led by Accel, Founders Fund and Factorial Funds in May. The jump follows an initial deal to sell Anthropic around $250 million of chips, a contract both sides intend to expand. Founded in 2022 by Oxford roboticist Walter Goodwin, Fractile builds hardware aimed at cutting the time models take to produce answers, though its chips are not expected to be in use until 2027.

05

Cognition CEO denies report that SpaceX tried to acquire the startup

TechCrunch AI ↗

Bloomberg reported that SpaceX approached AI coding startup Cognition about an acquisition, citing sources familiar with the talks. Cognition chief executive Scott Wu rejected the account on X, calling the story inaccurate, saying the company is not for sale and that the two firms never held discussions. Cognition raised $1 billion in late May at a $25 billion post-money valuation and is reportedly in early talks at $40 billion, while SpaceX has already bought xAI and paid $60 billion in stock for Cursor as it pushes into enterprise AI.

06

The U.S. banned Nvidia's best chips from going to China. Now it's trying to close a crucial loophole

CNBC Technology ↗

Chinese AI firms have been renting time on Nvidia's most advanced chips, including GB300s, through data centers in Southeast Asia, sidestepping export rules that bar them from importing the hardware. White House official Michael Kratsios accused Moonshot AI of tapping GB300s via a facility in Thailand in July. The House passed the Remote Access Security Act 369 to 22 to extend export-control logic to cloud GPU rentals, and lawmakers are now weighing customer vetting and location checks for hyperscalers without kneecapping US cloud providers abroad.

07

GLM-5.3 hits the API at $1.4/$4.4 per million tokens

VentureBeat ↗

Chinese startup z.ai put its new frontier model GLM-5.3 on its API at $1.40 per million input tokens and $4.40 per million output tokens, with cached input at $0.26 and cache storage free for a limited period. Artificial Analysis scored the model 60 on its Intelligence Index, seven points above GLM-5.2 and tied with Kimi K3 as the strongest open-weight model available, with sharp gains on agentic coding and cyber benchmarks including Terminal-Bench 3.0. Weights have been promised but not released, so for now the model is API-only rather than genuinely open source.

08

Nvidia has discussed an investment in Mercor as part of a round valuing the startup at $20B

TechMeme ↗

Nvidia has held talks to invest in Mercor, the marketplace that connects AI labs with human experts who label and generate training data, as part of a round valuing it at $20 billion. That would roughly double the $10 billion price Mercor set in its $350 million Series C in October 2025, and comes as its annualized revenue run rate doubled to $2 billion in four months. Nvidia paid Mercor tens of millions of dollars last quarter, and the talks appear strategic rather than part of a broader fundraise.

09

Google strikes $12bn AI chip deal with Marvell

Financial Times Tech ↗

Marvell issued Google a warrant to buy nearly 59 million of its shares at $206.58 each, an aggregate exercise price of roughly $12.2 billion, as part of an expanded custom silicon partnership. Marvell will build AI inference accelerators, networking chips, storage controllers and memory interface parts tied to Google's TPU ecosystem, with warrant tranches vesting against every $500 million of chips Google buys. Marvell shares jumped more than 11 percent in premarket trading while rival Broadcom fell over 3 percent, and full exercise would make Google the company's fifth-largest investor.

10

Flock Has a Powerful New AI Tool for Police. We Got Its Code

Wired ↗

Wired reporters and an independent security researcher reconstructed OS Investigate, an experimental Flock Safety tool now in testing with a small group of police agencies, using code left exposed on the company's login portal. The system lets officers search for people and vehicles by movement patterns alone, with no license plate, name or underlying crime required, and can cross-reference results against case files, 911 dispatch logs and commercial identity records to surface a person's name, home address, relatives and friends. Flock says the product is separate from its license plate readers, but it lands as the Washington Post reports at least 50 officials have been accused or charged over misuse of plate-reader data, with Flock cameras involved in 46 of those cases.

← 2026-08-18 All briefings 2026-08-20 →

Tools in this briefing, reviewed in our directory:

Kimi K3 → GLM-5.2 →

Track the tools behind the news:

→ AI Tool Leaderboard → Browse the directory