01
Bloomberg Technology ↗
The US Justice Department has spent nearly a year running a civil antitrust investigation into whether Andreessen Horowitz partners are improperly sitting on the boards of competing AI companies. At issue are Databricks, where co-founder Ben Horowitz holds a board seat, and Fivetran, where partner Martin Casado sits on the board, two firms that both help businesses collect, organize and analyze large troves of data. The inquiry tests Section 8 of the Clayton Act, which bars interlocking directorates between rival companies above certain size thresholds. It is the first time the agency has aimed that theory at a venture capital firm rather than private equity, and no enforcement decision has been reached.
02
CNBC Technology ↗
Alibaba released Qwen3.8-27B, a model small enough to run on a consumer laptop, and at the same time published the open weights for Qwen3.8 Max, its most powerful system. The company says the 27B model matches a model ten times its size on coding, professional tasks, research and multi-step agentic workflows. The launch lands one week after Meta open-sourced its strongest model and shipped laptop-class versions, putting the two rivals in direct competition for the same hardware. Hugging Face now counts 151,448 Qwen derivative models, roughly 2.6 times Meta's footprint.
03
Forbes Tech ↗
Wiz says its autonomous Red Agent found and exploited a script injection flaw in Snowflake's .NET connector repository that let unauthenticated users run arbitrary commands inside GitHub Actions runners simply by opening an issue with a crafted title. The exploit reached sensitive data in Snowflake's internal Jira environment, and the bug was mitigated on June 23 through Snowflake's bug bounty program. GitHub Copilot Autofix had approved the code change that introduced the flaw without flagging it, a gap between two AI systems that Wiz researcher Gal Nagli says shows frontier models can already exploit supply chain risks on their own.
04
Ars Technica ↗
A bookseller who suspected that a 1,000 book order placed through Biblio was really destined for an AI company hid a tracking tag in the shipment, and reporters followed it to an Amazon warehouse in Las Vegas run by a team called VGT3 whose logo is a Tyrannosaurus rex holding a book. Workers there say spines are cut off the books to speed up scanning, destroying the copies in the process. Amazon buys printed books in bulk and uses the scans as training data for its Nova models, and printed texts are prized because they often do not exist online and predate 2022, so they are free of AI-generated content. An Amazon spokesperson said the company buys books through commercial channels to improve its products.
05
TechCrunch AI ↗
Groq raised a $350 million Series A at a $3.5 billion valuation led by Disruptive, with Nvidia expected to participate. That is far below the $6.9 billion Groq commanded in September 2025, a drop that followed Nvidia's $20 billion licensing deal that took founder Jonathan Ross and other key staff, though the company frames the round as a new baseline rather than a down round. Having stepped back from its proprietary inference chips, Groq now runs 13 data centers on Nvidia GPUs serving more than 6 million developers and enterprises, and plans to grow from 54 megawatts to over 200 megawatts by 2027. CEO Alex Davis said the goal is to build Groq into the world's leading AI inference cloud.
06
Financial Times Tech ↗
Higgsfield raised $400 million in a Series B at a $5.4 billion valuation, with Goldman Sachs and Intel joining DST Global, Liberty Global, Tribe Capital, Smash Capital, Fifth Wall and NTT DOCOMO Ventures. That is more than four times the $1.3 billion the AI video start-up was worth earlier this year, when it raised $80 million. Founded by former Snap executive Alex Mashrabov, the company says annualized revenue reached $700 million in August, up from $20 million a year earlier, as it targets marketers and social media creators against rivals such as Runway and Synthesia.
07
Unite.AI ↗
Gravis Robotics raised $200 million in a Series A led by SoftBank, billed as the largest Series A in construction robotics so far. The Zurich company, an ETH Zurich spinout founded in late 2022 by CEO Ryan Luke Johns and CTO Dominic Jud, does not build machines but retrofits contractor-owned excavators and earthmovers with a modular rack of cameras, lidar, satellite positioning and edge computing. Its learning-based control system blends machine hydraulics data with those sensors and adapts to different machines without being programmed for each one. Systems are already running in seven countries, including autonomous excavation at Manchester Airport and quarry work, with partners including Holcim, Taylor Woodrow, HD Hyundai and Flannery Plant Hire.
08
IEEE Spectrum AI ↗
A team at Axiom Math has used its automated prover to verify the formal proof of the so-called 246 theorem, which states that infinitely many pairs of primes differ by no more than 246. That bound is the closest mathematicians have come to the twin prime conjecture, built by Fields Medalists James Maynard and Terence Tao on top of Yitang Zhang's 2013 breakthrough. The machine-checked Lean 4 formalization was published on August 17 with 41 named contributors, led on the human side by Carnegie Mellon doctoral student Sidharth Hariharan. Axiom founding mathematician Ken Ono argues the real payoff is using the same machinery to verify AI-generated code in critical infrastructure, finance and cybersecurity.
09
Unite.AI ↗
Researchers from Japan's National Institute of Advanced Industrial Science and Technology, the University of Tsukuba, the University of Technology Nuremberg and the University of Oxford found that changing nothing but text color shifts how vision language models read meaning. Testing four open-source models, they measured sentiment swings of up to +0.42 when positive words were colored green or blue and -0.48 when negative words were colored red, with Qwen2-VL-7B the most susceptible and IDEFICS2-8B showing a 0.52 total range. In visual question answering tests, highlighting a wrong decoy answer through contrast alone pushed error rates up, and the effect grew as the surrounding text was made harder to see. The authors call visual styling a critical and previously underexplored vulnerability in model pipelines.
10
TechCrunch AI ↗
Relay, an AI workflow automation tool launched in 2021 to compete with Zapier on jobs like document drafting, copyediting and project management, is shutting down. Founder and CEO Jacob Bank is joining Google as a VP of product on the Chrome team, returning to a company where he spent more than six years after Google acquired his scheduling app Timeful in 2015. Free customers lost access on August 15 and paying customers will be cut off on September 14. Bank said he has ambitious plans to help people work with AI in Chrome to get things done.