Sakhanda Wire
NVDA $150.96 -2.89% MSFT $508.29 +0.40% GOOGL $282.44 +0.80% META $616.38 -0.40% AMZN $267.74 +3.82%

Latest news

Older stories

The Verge

More than one million users clicked LinkedIn’s AI slop button

More than one million people have clicked LinkedIn’s “Seems like AI slop” button, according to chief product officer Hari Srinivasan. LinkedIn introduced the control on July 30, and users can find it in the three-dot menu attached to a post.

MarkTechPost

Anthropic Adds Claude Mythos 5 to Enterprise Security Scans

Anthropic has moved Claude Mythos 5 into Claude Security scans, which are available in public beta to Claude Enterprise customers from August 21, 2026. Organizations do not need a separate model add-on: usage is billed as standard tokens under their existing plan. An administrator enables the feature, and users launch scans at claude.ai/security. The service is limited to Enterprise organizations with code stored in GitHub; Pro, Max, and Team customers are not eligible for the Mythos-powered scans.

TechCrunch

Nvidia research highlights the importance of AI agent harnesses

New research from Nvidia suggests that the software harness surrounding an AI model can matter more than the model itself when agents handle long-horizon tasks. A harness manages memory, context, feedback, tools, runtime and related libraries. Nvidia researchers used a custom system with a memory-focused design and a supervising component to help Claude Opus 5 achieve a 100% score on ARC-AGI-3, an interactive reasoning benchmark based on instruction-free 2D games. Without that harness, the model scored 30%, the highest result among the models tested.

The Decoder

Anthropic deploys Claude Mythos 5 for cybersecurity defense

Anthropic has begun operating its Claude Security code scanner on Claude Mythos 5, the company’s most capable model. The scanner is available in public beta to Enterprise customers and examines codebases for vulnerabilities. Its reports assign each finding a severity rating and a CWE classification, then propose a possible fix. Scans are billed as regular token usage, and every patch must still be approved by a human.

The Decoder

Deepseek releases experimental multimodal vision model

Deepseek has released Deepseek-V4-Flash-Vision-Exp, an experimental multimodal model that adds image processing to the text capabilities of Deepseek-V4-Flash. The company says the vision version preserves the base model’s performance in reasoning and world knowledge. On Deepseek’s internal multimodal agent benchmarks, it scores close to Opus 4.8 and sometimes exceeds it.

The Decoder

Survey Finds Sharp Rise in US Opposition to Nearby Data Centers

Seventy-five percent of Americans now oppose building a data center near where they live, according to a Heatmap News survey. The figure includes 61 percent who said they were strongly opposed. One year earlier, views were almost evenly divided: 43 percent supported nearby construction and 42 percent opposed it. A majority against data centers first appeared in February.

The Decoder

US weighs asking partners to choose between Washington and Beijing on AI

The United States is preparing a letter urging partner countries to take a side in the competition between Washington and Beijing over artificial intelligence, Reuters reported. A State Department draft reviewed by the agency says countries that join China’s AI initiative would be excluded from the US-led Pax Silica coalition. The draft warns that trying to participate in both frameworks means belonging to neither.

TechCrunch

DOJ Investigates a16z Board Seats at Competing Companies

The U.S. Department of Justice has reportedly spent almost a year investigating a board arrangement involving venture firm Andreessen Horowitz, also known as a16z. The firm has two partners serving on boards of companies that now compete in overlapping markets: Ben Horowitz sits on Databricks’ board, while Martin Casado serves on Fivetran’s board.

TechCrunch

Starcloud raises $250 million to expand orbital data center plans

Starcloud has raised an additional $250 million, extending its $170 million Series A round from March. The financing values the company at $2.3 billion and will support a larger manufacturing site as well as development of Starcloud-3, its biggest planned orbital data center spacecraft. The vehicle is intended to launch on SpaceX’s Starship.

The Verge

YouTube filmmakers face backlash over videos promoting AI tools

Prominent filmmaking creators Matti Haapoja and Sam “Kold” Kolder are facing backlash after posting videos that showcase Higgsfield and its recently added Seedance 2.5 functionality. The videos present the services as tools that could significantly change entertainment production, prompting criticism over creators accepting money connected to AI promotion.

DeepMind

Google DeepMind expands game-based AI research with EVE partnership

Google DeepMind says it is working with game developers to create playable prototypes that explore new forms of gameplay and advance AI research. The company describes games as a long-standing testing ground for systems including DQN, AlphaGo, AlphaZero, MuZero and AlphaStar, whose capabilities later informed work such as AlphaFold.

The Decoder

Waymo develops custom chip for its latest robotaxi vehicles

Waymo, Alphabet’s robotaxi subsidiary, has developed a custom chip for self-driving vehicles, Bloomberg reports. The component is already operating in the company’s latest generation of robotaxis. Waymo says it can process sensor data more quickly and run the artificial-intelligence models used to interpret the surrounding environment and guide the vehicles’ responses.