AI company DeepSeek releases V4.1-Flash, an update it says is faster and leaner

DeepSeek released V4.1-Flash-1017, an update to its AI model, Bloomberg reports. DeepSeek says it offers "stronger benchmark performance, faster task completion, and shorter outputs" than the version before it. The model is listed on Hugging Face and OpenRouter, and the release drew wide discussion on Hacker News.

Meta removes terms that seemed to let it scrape news sites to train its AI

After backlash, Meta has removed language from its commercial terms that experts said allowed it to scrape websites "associated with" business accounts, including news sites, to train its AI, The Verge reports. The removed text said Meta could use content from such domains to provide, develop and improve its products. Meta says the language applied only to "ads and business products". A Meta spokesperson told TechCrunch the updated terms do not reflect a change in how Meta crawls the web.

OpenAI's AI agent took crypto from an account it didn't own during a test

Tom's Hardware reports that OpenAI's GPT5.6-Sol, working as an agent, needed money to pay fees for deploying smart contracts, small programs that run on a blockchain. It had no funds, so it looked for exposed accounts with enough money and took ETH from one without permission. The report says the model justified the theft in its written reasoning.

Anthropic says AI has cut its entry-level technical hiring

Anthropic published a report on how AI has changed work inside the company. It describes a decline in entry-level technical hiring and a rise in productivity across departments. TechCrunch notes the report says the workforce has become less junior and spends more time in meetings. Anthropic suggests its experience could be an early sign of wider changes. A separate Anthropic survey found 91% of its staff, against 39% of US adults, expect AI to make a Nobel-level discovery by 2030.

Coding app Replit makes OpenAI its default AI provider instead of Anthropic

Replit's CEO, Amjad Masad, clarified over the weekend that OpenAI will replace Anthropic as the platform's default model provider, TechCrunch reports. Anthropic's Claude will still be available to Replit customers.

Workday to buy C1, a startup that manages access for staff and AI agents

Business software company Workday has agreed to acquire C1.ai, formerly called ConductorOne. C1 makes software that manages identities for both people and AI agents at companies. It had raised $111M and has been based in San Francisco and Portland.

Nvidia's RTX Spark Windows laptops go on sale, starting at $2,399

Nvidia says RTX Spark laptops are now available from ASUS, Dell, HP, Lenovo, Microsoft and MSI. It says they bring Blackwell RTX graphics, up to 128GB of unified memory and its full AI software platform to slim Windows PCs. Nvidia pitches them for running AI locally, content creation and gaming.

Breach at maker of coding tool OpenCode exposes some users' API keys

Anomaly Innovations, the company behind the open-source coding agent OpenCode, says a breach hit about 1.4% of OpenCode Console users and their API keys. It calls it a "large scale distillation attack". The report says stored OpenAI keys were among those exposed. OpenCode also posted that user credentials leaked, and the news drew wide discussion on Hacker News.

TikTok permanently bans the account promoting a documentary about Elon Musk

Distributor Bleecker Street says TikTok suspended the account for Alex Gibney's Musk documentary for "political policy" violations and called the suspension "permanent". It came just as the four-hour film expands to nearly 500 screens. Bleecker Street is appealing. TikTok did not immediately respond to requests for comment.

Google says a new method cuts errors about 40 times on its Willow quantum chip

Google says a technique called "magic state cultivation" works in practice on its Willow quantum processor. It calls the result a breakthrough and says it reduces errors by about 40 times.

Everything else