Anthropic Launches Public-Benefit Fund With 5% Equity Stake

Morning, {{first name | folks}}! AI is advancing on several fronts this week, from Anthropic’s new public-benefit fund to coding agents speeding up models and DeepSeek opening web search to developers. Meanwhile, Meta’s revised terms and a major vendor security incident raise questions about content access and digital trust.

Anthropic is committing 5% of its shares to a new Public Benefit Fund, with the board able to allocate another 2% for every 50% increase in the company’s value above a set baseline. The structure gives the fund the potential to grow alongside Anthropic’s valuation.

The fund could support nonprofits, direct payments, and programs addressing AI’s impact on workers. Combined with existing founder and employee commitments, Anthropic says roughly 20% of its shares will be dedicated to public benefit, extending the company’s stated mission beyond AI development itself.

vLLM used AI coding agents to make Kimi K3 run faster on NVIDIA’s GB300 GPUs. The changes delivered up to 2× higher throughput in tested setups, meaning the system could generate more tokens in the same amount of time.

The project also highlights a growing challenge for AI-assisted engineering: reviewing the code agents produce. vLLM says the volume of generated code is outpacing manual review, pushing the team to rely more on automated tests to check correctness and performance before changes are used in production.

DeepSeek has opened its own search engine to developers through two new APIs: one finds relevant web pages, while the other pulls their content into AI workflows. Each costs $0.75 per 1,000 successful requests, giving developers another option for building AI agents that can search the web and read source material.

The launch comes alongside DeepSeek-V4.1-Flash-1017, an updated model that the company says completes tasks faster and produces shorter outputs than its predecessor. DeepSeek is also pitching the model for building entire projects from scratch, from 3D scenes to playable games.

Meta’s new commercial terms are raising concerns among news publishers over the company’s ability to scrape content from websites linked to business accounts. Set to take effect on October 30, the rules could give Meta broad permission to collect news articles and other website content, legal experts warn.

Meta has removed explicit references to AI training and narrowed the stated uses to advertising and business products. But legal experts still question whether the revised wording leaves room for content to be collected for other purposes. The distinction matters to publishers already grappling with how AI companies access their work to develop and power their products. Meta says its crawling practices aren’t changing.

A compromised ConnectWise Cloud Backup credential gave attackers access to Microsoft 365 environments across multiple organizations. Huntress detected more than 300 successful sign-ins across nearly 1,000 tenants, potentially exposing emails, files and sensitive business data.

Attackers also used existing app permissions to expand their access, turning one compromised vendor credential into a wider security concern. The incident highlights how third-party tools can create risks across multiple customer environments when their credentials are compromised.