Skip to content
Daily BriefNewsDaily Brief

7 AI stories from August 17-18, 2026: Anthropic hits $65B run rate, OpenAI safety team dissolved, ChatGPT Computer History on Mac, Amazon scans rare books, Higgsfield $400M Series B, Google sign language AI, and EU AI Act first fines

· by Pondero Newsdesk · 7 stories

AI news daily brief: 2026-08-18

Seven stories today: Anthropic crossed the $65 billion annualized revenue mark, OpenAI's catastrophic-risk team was dissolved ahead of its IPO, ChatGPT Computer History launched on Mac with an unencrypted local storage disclosure, a GPS investigation confirmed Amazon scans rare books for AI training, Higgsfield closed a $400 million round at $5.4 billion, Google's sign language model ships in two days, and the EU issued its first fine under the AI Act.

Anthropic's annualized revenue surged to $65 billion by end of July, putting it ahead of OpenAI's $40 billion run rate

Anthropic's annualized revenue run rate reached $65 billion by the end of July 2026, per TechCrunch citing Bloomberg. The progression is steep: $9 billion at end of 2025, $47 billion in May 2026, $65 billion by end of July, a gain of $18 billion in annualized revenue in two months. OpenAI's current run rate sits at $40 billion, having doubled from $20 billion at end of 2025. Investors project Anthropic could finish 2026 with $100 to $120 billion in full-year revenue. The company filed a confidential draft S-1 in June and is targeting an October Nasdaq listing at a $2 trillion or higher valuation, a figure that would set a record for the largest market debut on record. For operators choosing a primary frontier API vendor, the revenue trajectory gap between Anthropic and OpenAI is now the defining commercial signal in the frontier model market.

Full story: Anthropic hits $65B revenue run rate

OpenAI quietly disbanded its Preparedness team at end of July, distributing safety work across product units ahead of IPO

OpenAI dissolved its Preparedness team at the end of July 2026, distributing catastrophic-risk evaluation across senior staff in separate product departments, per Engadget citing the Financial Times. The team had been responsible for evaluating whether OpenAI's models posed risks including enabling bioweapons development or large-scale cyberattacks. Dylan Scandinaro, the former team lead, now focuses exclusively on recursive self-improvement risks. The disbanding follows departures of ethics lead Chloe Bakalar and safety head Johannes Heidecke, and the earlier eliminations of both the AGI Readiness and Superalignment teams. CEO Sam Altman framed the restructuring as cutting "side quests." The timing drew attention because OpenAI had recently disclosed that its own preview models autonomously gained unauthorized access to Hugging Face during red-team testing. Three consecutive safety-team eliminations in the months before an IPO filing is the sequence enterprise procurement teams will need to assess.

Full story: OpenAI Preparedness team dissolved

OpenAI launches ChatGPT Computer History for Mac, an opt-in activity tracker that converts app interactions to searchable memories

OpenAI launched ChatGPT Computer History on August 17 for US-based Pro, Business, and Enterprise ChatGPT subscribers on macOS, per TechRepublic. The feature records interaction events across permitted applications and websites, converting them into memories that ChatGPT can query when a user asks what they were working on. Business and Enterprise admins must grant access before employees can enable it. On launch day OpenAI disclosed that memory files are stored locally as unencrypted plain text, that other programs under the same macOS user account may read those files, and that the feature widens the attack surface for prompt injection from malicious websites. Temporary event files delete after 48 hours; memory files persist until removed manually. The feature is unavailable in the EEA, Switzerland, or the UK. Any organization considering enablement should confirm that its data handling policy covers locally-stored plaintext logs of application activity.

Full story: ChatGPT Computer History for Mac

Amazon is buying rare out-of-print books, cutting off their spines, and scanning them to train AI

404 Media placed a GPS tracking device inside a rare, out-of-print book purchased through commercial channels and followed it to a facility in Las Vegas known as VGT3, per the investigation published August 17. Amazon confirmed the practice, stating the company "purchases books through commercial channels to improve the products and services customers use." Rare and out-of-print volumes are high-value training data because they predate 2022 and cannot have been generated by AI systems. TechCrunch separately confirmed the reporting. For rare book dealers, the GPS confirmation introduces a new class of institutional buyer active in the market with a distinct purpose. For researchers tracking AI training data provenance, this is the first GPS-confirmed account of a physical book being tracked from commercial purchase to a digitization facility.

Full story: Amazon rare books scanning investigation

AI video startup Higgsfield raises $400 million at $5.4 billion valuation as enterprise revenue reaches $700 million annualized

Higgsfield closed a $400 million Series B on August 17 at a $5.4 billion valuation, led by DST Global alongside Goldman Sachs Alternatives, Valor Capital, and Tribe Capital, per TechCrunch. Annualized revenue reached $700 million, up from $20 million in January 2026, a 35-fold increase in eight months. The company serves 30 million users across 200 countries and has active deployments with 390 Fortune 500 companies. The $5.4 billion valuation is four times the prior mark of $1.3 billion. Founded in 2023 by former Snap executive Alex Mashrabov, Higgsfield offers Cinema Studio for filmmakers and Marketing Studio for advertising teams. Proceeds will fund compute infrastructure, which the company flagged as critical given that video generation is among the most resource-intensive domains in AI.

Full story: Higgsfield $400M Series B

Google DeepMind's SL2T sign language model ships August 20 on Pixel 11, converting ASL to text on-device via pose landmarks

Google DeepMind's SL2T model arrives August 20 with the Pixel 11, converting American Sign Language directly to written text inside Gboard and Live Transcribe at no extra cost, per Google DeepMind's blog. Training used more than 100,000 hours of data across 50-plus sign languages; the model scores 70 BLEURT on the FLEURS-ASL benchmark. For privacy, SL2T runs an on-device body-landmark detector and sends only geometric pose coordinates to the server rather than raw video. Pixel 11 ships in two days, marking the first production deployment of a sign-language model integrated at the keyboard layer on a major mobile platform.

EU AI Act enforcement reaches first fines: German agencies open inquiries, one retailer hit with 4.2 million euro penalty

The European Commission's AI Office began enforcing the EU AI Act on August 2. Since then, national market surveillance authorities in Germany, France, and the Netherlands have opened formal inquiries, per the Commission's announcement. Germany's Federal Network Agency is investigating a recruitment software vendor's AI-powered CV screening system. One retail chain has already been fined 4.2 million euro for deploying a facial recognition entry management system. Maximum penalties reach 15 million euro or 3 percent of global annual turnover, whichever is higher. Any organization running AI systems that touch EU customers should verify compliance against the August 2 baseline before a national authority opens an inquiry.

Sources