AI News and Industry Updates

One Local Workstation, a 27B Model, Half an Hour: A Commercial App's License Check Taken Apart by Static Analysis

XDA lead technical editor Adam Conway published a hands-on test on August 22: running Q8-quantized Qwen 3.8 27B on an Nvidia GB10 workstation with 128GB of unified memory, he asked it to analyze the license check in a commercial application he had bought, entirely offline and without ever executing the program. In roughly 30 minutes the model disassembled ARM64 code, worked through thousands of lines, recovered an RSA public key the vendor had deliberately obscured, mapped the signature check, machine binding and revocation list, caught its own mismatched hash mid-way and corrected it, and produced a proof-of-concept script. Conway does not name the software, stresses that it was a legitimately purchased copy, and frames the point as capability moving into models small enough to run locally.

AI Content Editor1 weeks ago4 views

The New MCP Roadmap Drops the Request-Response Assumption: Server-Initiated Events, Tools Revealed on Demand, Agents With Workload Identity

MCP lead maintainers David Soria Parra and Den Delimarsky published a new roadmap on August 22 listing five priorities: agentic messaging primitives (server-initiated events, webhooks and channels so clients stop polling, with the Tasks extension SEP-2663 graduating from experimental), unifying on HTTP-native transport, agent identity and enterprise security (DPoP, workload identity federation, standardized delegation), improved primitives (a single tool-result contract plus progressive tool discovery) and SDK developer experience. The roadmap says SEPs falling inside these areas get expedited review, and publishes no delivery dates at all.

AI Content Editor1 weeks ago9 views

Anthropic Launches Claude Academy: 26 Free Courses Reorganized Around Products, With Certificates and an Architect Exam

Anthropic launched Claude Academy on August 21, moving the former Anthropic Academy from anthropic.com/learn to academy.claude.com. The catalog grew from 13 to 26 courses, restructured into four learning paths built around actual products — Claude Code, API integration, MCP and agent development — plus topics like prompt engineering and RAG. The three API courses each run past 60 lessons and take eight or nine hours. Signup needs only an email, everything is free, and completed courses grant certificates. Anthropic also introduced a Claude Certified Architect exam for people designing real-world applications.

AI Content Editor1 weeks ago14 views

Google Turns "Add to Preferred Sources" into Two Lines of JavaScript Publishers Can Embed — After AI Summaries Took the Clicks

Google introduced an embeddable Preferred Sources button on August 20: publishers add two lines of JavaScript to an article page and get a localized, Google-styled button. A reader taps it, confirms with "Add" on a short Google screen, and lands back on the same article — after which that site appears more often in Search, Discover and Google News. Google says more than 345,000 unique sources have been selected since the feature launched, up from roughly 90,000 in December 2025 and 200,000 in April 2026. Preferred Sources has been a ranking signal since mid-2025 and reached every language in April, but it does not override relevance.

AI Content Editor1 weeks ago2 views

Linear's Data: AI Now Writes Nearly Half of All Issues and Teams Using Coding Agents Tripled Their PRs — but Time Spent Coordinating Went Up Too

Linear published an AI usage report built on aggregated data from its paid workspaces, covering June 2024 through August 2026. Two years ago fewer than one issue in a thousand was created by AI; now it is close to half. Teams using coding agents went from 21 to 65 PRs per week, while teams without them moved from 8 to 10. Running the other way: between June 2025 and June 2026, time spent creating, triaging and commenting rose across nearly every function — engineering added about 5 minutes a month on create and triage alone, and founders added 26 minutes on commenting. Product managers attaching PRs rose from 3% to 10%, designers from 1% to 8%.

AI Content Editor1 weeks ago4 views

Apple Music Will Label AI-Generated Songs, Moving from Optional Disclosure to Required — Without Saying How It Will Check

Per an email obtained by Billboard, Apple Music told labels and distributors on August 20 that later this year, any song its content providers tag as "materially generated using AI" will carry a visible label on the service. Compared with the optional AI Transparency Tags introduced in March, the wording hardens to a requirement whenever AI created a material portion of the content, with providers responsible for the call because they best know how their content was made — but the email gives no enforcement detail. An Apple executive said earlier that more than a third of monthly uploads to Apple Music are 100% AI-generated, while AI music accounts for under 0.5% of actual listening.

AI Content Editor1 weeks ago4 views

Gemma Passes 1 Billion Downloads and 100,000 Community Variants, the First Cumulative Adoption Figure Google Has Given for Its Open Models

Google announced on August 20 that its Gemma family of open models has passed a billion cumulative downloads, with developers publishing more than 100,000 variants over two years — the first cumulative total Google has attached to Gemma since the family launched in early 2024. Deployments highlighted include NASA's Jet Propulsion Laboratory flying a 4-bit compressed Gemma 3 4B on a Loft Orbital satellite, India's National Health Authority wiring Gemma 4 into a health app with over 100 million Android downloads, and Yale and Google's C2S-Scale identifying a cancer therapy pathway later verified in living cells.

AI Content Editor1 weeks ago6 views

GitHub AI Weekly · Issue 02

GitHub AI Weekly · Issue 02 (Aug 22, 2026) · AI Beyond

This issue covers the GitHub Trending weekly board for week 34 of 2026: the top 10 AI projects by stars gained this week, spanning multimodal, AI coding, RAG, agents, inference, and infrastructure. Each entry states what changed this week, where the project actually fits, and what to know before installing it, with deeper reporting on three of them. Data as of Aug 22.

AI Content Editor1 weeks ago16 views

Nvidia Pays $6 Billion to License Poolside's Model Factory and Offers Jobs to 109 Staff — and Both Sides Say It Is Not an Acquisition

Per an investor letter obtained by Newcomer, Nvidia struck a non-exclusive $6 billion licensing deal for Poolside's Model Factory — the infrastructure suite covering data processing, training, reinforcement learning and evaluation that Poolside uses to build its Laguna family of open-weight coding models — plus a $1 billion investment at a $12 billion pre-money valuation. Nvidia will extend offers to 109 employees who worked on the model; Poolside's three co-founders stay. Poolside plans to distribute the $6 billion to investors by the end of next year.

AI Content Editor1 weeks ago8 views

Google Gets a $12.2 Billion Marvell Warrant: No Cash Up Front, One Tranche Unlocked per $500 Million of Chips Bought

A regulatory filing disclosed on August 19 shows Marvell issued Google a warrant exercisable at $206.58 per share for up to 58,970,907 shares — roughly $12.2 billion, or about 7% of Marvell's shares outstanding — running through August 18, 2033. Google pays nothing up front: 240 equal tranches vest one at a time for each $500 million in custom products revenue Marvell books from Google's orders. The underlying commercial agreement, signed July 29, covers silicon around Google's TPU ecosystem — inference accelerators, storage and network controllers — but not the TPUs themselves.

AI Content Editor2 weeks ago3 views

Ray's CVSS 9.4 Flaw Is Confirmed Under Active Exploitation: Visiting One Malicious Page While Running Ray Can Execute Code

CISA added Ray's CVE-2025-62593 to its Known Exploited Vulnerabilities catalog on August 17, giving federal agencies until August 20 to upgrade or stop running it — one of the tightest windows it issues. The flaw scores CVSS 9.4: before version 2.52.0, browser protection on the Ray dashboard and API checked only whether the HTTP User-Agent began with "Mozilla," which combined with DNS rebinding allows arbitrary code execution on a developer's machine and lets the browser pivot to network-adjacent Ray instances. Bitsight reports the RondoDox botnet began exploitation attempts on November 24, 2025.

AI Content Editor2 weeks ago4 views

Given Only a Paper's Pre-Publication Bibliography, Frontier Models Recover Its Original Idea 3–15% of the Time

A new benchmark, Reconstruction (arXiv:2608.16645), hands a model only the literature that existed before a paper was published — with author names and metadata stripped — and asks what that paper proposed. Across six scientific domains and 643 papers, seven frontier models reach Match rates of roughly 3–15%; a reference-only multi-agent pipeline with no web search pushes that to about 42%. The authors note that many "AI scientist" evaluations let models see full paper text, author information or post-publication signals, so high scores under those conditions cannot be read as genuine inferential capacity.

AI Content Editor2 weeks ago3 views

Showing 12 of 12 stories