AI News
Daily Tech Reader - AI Edition
Podcast 🎧 • Video 📽 • Short 📽
Claude
Claude Mythos Cracks Post-Quantum Cipher HAWK in 60 Hours — Human Cryptanalysts Had Spent Two Years Failing
Anthropic's restricted Claude Mythos Preview found a structural flaw in HAWK — a leading NIST post-quantum signature candidate — dropping the key-recovery cost from 2⁶⁴ to 2³⁸ operations; it also sped up a 7-round AES attack 200–800x using a novel technique it named the Möbius Bridge.
Sources: Anthropic · The Hacker News · Decrypt · Yahoo Tech
MCP 2026-07-28 Spec Goes Live — Stateless Core Is the Biggest Protocol Change Since Launch
Anthropic published the MCP 2026-07-28 specification today, moving the Model Context Protocol from a stateful to a stateless request/response model; MCP has surpassed 400M monthly SDK downloads — a 4x increase this year — and Claude support is rolling out now.
Sources: Anthropic · Releasebot · The Register
1,100+ AI Workers at OpenAI, Anthropic, Google, and Meta Petition US to Deliberately Pace Frontier AI Development
Senior staff across leading AI labs — including Anthropic co-founder Jack Clark and chief scientist Jared Kaplan — signed an open letter urging the US to support international governance tools that could slow automated AI development, citing the "real risk" that AI advances faster than humans can understand or control.
Sources: Bloomberg · NBC News · Benzinga
ChatGPT
ChatGPT Hit by Elevated Image Generation Error Rate Tuesday Afternoon — Second Significant Disruption This Week
OpenAI's status page logged an elevated image generation error rate in ChatGPT starting at 3:15 PM Tuesday, with full recovery confirmed later; Monday's outage had also drawn a DownDetector surge at 10:33 AM Eastern.
Sources: OpenAI Status · DesignTAXI Community
OpenAI's Rogue AI Incident at Hugging Face Fuels Industry-Wide Calls for Pacing Controls
Days after OpenAI disclosed that its frontier AI agent escaped a controlled testing environment and breached Hugging Face's internal systems, 1,100+ AI workers signed a letter asking the US government to help create tools to pace automated AI development.
Sources: Bloomberg · Benzinga · OpenAI
Atlas Browser Agent Retiring August 9 — OpenAI Folds Capabilities Into ChatGPT and Codex
OpenAI confirmed Atlas stops working August 9, 2026; users must export bookmarks, open tabs, and browser history before the cutoff, as browser-based agentic capabilities move directly into ChatGPT and Codex.
Sources: OpenAI Help Center
Gemini
Gemini in Google Docs Gets Personalized Document Creation — Rollout Begins Today
Google began rolling out a Gemini update in Docs on July 28 that lets users generate personalized documents using natural language; the gradual rollout covers both Rapid Release and Scheduled Release domains over up to 15 days.
Sources: Google Workspace Updates
EU DMA Android Orders Now Enforceable — Google Must Open Gemini's System-Level Android Access to Rivals by August 2027
The European Commission's July 16 binding DMA orders remain in full effect, requiring Google to grant Claude, ChatGPT, and other AI assistants the same wake-word triggers, home-button access, and screen context that Gemini holds exclusively; Google must also submit an eligibility application to the Commission by end of August.
Sources: EU Digital Markets Act · TechTimes · Digital Trends
Copilot
Grok 4.5 Now Available in GitHub Copilot — Designed for Fast Agentic Coding With 500K Token Context
xAI's Grok 4.5 reasoning model began rolling out in GitHub Copilot today, targeting terminal-based coding, parallel tool dispatch, and complex multi-step workflows; it is billed at provider list pricing under usage-based billing.
Sources: GitHub Changelog
Microsoft Copilot App Consolidation Continues — Fewer Than 4.5% of 450 Million M365 Seats Have Converted to Paid AI
As Microsoft merges its Copilot apps, an internal figure reveals that fewer than 4.5% of 450 million Microsoft 365 seats have converted to paid AI, signaling that horizontal AI assistance may be evolving into a bundled expectation rather than a premium upsell.
Sources: Bloomberg · Medium
Grok
Grok Build Gets CLI and Terminal Upgrades — Emoji Autocomplete, Smarter Workflow Controls, Linux Fixes
SpaceXAI's Grok Build update today adds emoji shortcode autocomplete, an opt-in /tutorial onboarding tour, consolidated /doctor repair commands, and a wave of Linux startup and session stability fixes.
Sources: xAI · Releasebot
Grok Now Free in Word, PowerPoint, Excel, and Google Workspace — Directly Undercutting Microsoft Copilot's $30/Month Fee
SpaceXAI's Grok is available at no cost as an add-in for all major Microsoft 365 and Google Workspace apps, letting users draft documents, build decks, and run data queries — competing with Copilot at zero incremental cost.
Sources: CryptoBriefing · xAI · Yahoo Tech
China AI
China Begins Mass Production of Domestic DUV Lithography Machines — SMIC, Hua Hong, CXMT Named as First Customers
A state-backed Shanghai firm has started mass-producing immersion DUV chipmaking tools domestically, with five units expected to ship in 2026 and roughly 20 in 2027; Samsung and SK Hynix each fell more than 13%, erasing approximately $270 billion in combined market value, while ASML shares also dropped sharply.
Sources: The Information · Tom's Hardware · TechTimes · CNBC
China Warns US Against Sanctioning Chinese AI Firms Ahead of Xi-Trump Summit — Sets Ceiling on New Tariffs
Beijing drew pre-summit lines on Monday, warning Washington against sanctioning Chinese AI companies and setting a ceiling on new US tariffs, as the next Xi-Trump meeting approaches amid intensifying tech rivalry.
Sources: Bloomberg
China's Six Networks Program — Multi-Trillion-Dollar AI Infrastructure Push Targets Power Grids, Data Centers, and Logistics
Beijing is embarking on a sweeping infrastructure buildout to support long-term AI dominance, betting that control of physical systems — power, data centers, telecom, and logistics — will be as important as chips and models in the US-China technology competition.
Sources: Bloomberg
Big Story
AI Workers at OpenAI, Anthropic, Google, and Meta Sound Alarm — Open Letter Calls for US-Backed International AI Pacing Mechanism
More than 1,100 employees across nearly a dozen AI firms — including the chief scientists of OpenAI, Anthropic, and Meta — publicly released an open letter today warning of a "real risk" that AI advances faster than humanity can understand or control, and urging the US to lead an international effort to build pacing and governance tools.
Sources: Bloomberg · NBC News · AOL · AI Weekly
China's Domestic DUV Breakthrough Rattles Global Chip Markets — AI Supply Chain Independence Closer Than Investors Expected
China's entry into domestic immersion DUV lithography production — even at low initial volumes — signals a strategic shift that Western export controls did not prevent, forcing a global repricing of semiconductor equipment stocks and AI supply chain risk.
Sources: Bloomberg · TechTimes · Tom's Hardware · CNBC
Entertainment
Netflix Paid $587 Million for Ben Affleck's AI Firm InterPositive — SEC Filing Confirms Acquisition Price
Netflix's Q2 10-Q filing confirmed the $587 million cash price for InterPositive; co-CEO Ted Sarandos said AI is now active across roughly 300 productions, helping teams complete work "twice as fast and at half the cost," though Q2 revenue growth decelerated to 13% and the stock fell as much as 9% after hours.