AI News for April 4, 2026 — 7min.ai

[NEWS] Anthropic cuts off OpenClaw from Claude subscriptions, forces pay-as-you-go: Anthropic will block Claude Pro and Max subscribers from using their subscription limits with third-party tools like OpenClaw starting April 4 at 12pm PT.
[NEWS] OpenAI reshuffles leadership: COO steps aside, AGI chief takes medical leave: OpenAI COO Brad Lightcap is moving to a "special projects" role reporting to Sam Altman, focused on enterprise software partnerships with private equity firms.
[NEWS] DeepSeek v4 will reportedly run entirely on Huawei chips, ditching Nvidia: DeepSeek v4, expected to launch in the coming weeks, will run entirely on Huawei's Ascend chips.
[NEWS] Anthropic acquires biotech startup Coefficient Bio for $400M in stock: Anthropic purchased Coefficient Bio, an eight-month-old AI pharma startup with fewer than 10 employees, for roughly $400M in shares.
[TOOLS] Cursor 3 launches agent-first interface with parallel AI fleets: Cursor released version 3, a complete redesign built around AI agents that write most of the code.
[NEWS] Utah allows AI chatbot to prescribe psychiatric medications in US first: Utah approved a one-year pilot letting Legion Health's AI chatbot renew prescriptions for 15 lower-risk psychiatric medications including Prozac, Zoloft, and Wellbutrin.
[NEWS] OpenClaw vulnerability rated 9.8 severity gives attackers full admin control: OpenClaw developers patched three high-severity vulnerabilities, including CVE-2026-33579 rated up to 9.8/10.
[NEWS] Meta Superintelligence Labs quietly builds dedicated hardware team: Meta's Superintelligence Labs (MSL) is building a hardware team and hired Rui Xu, former hardware lead at AI agent startup Dreamer, to run it.
[NEWS] IWF documents 260-fold increase in AI-generated child sexual abuse material: The Internet Watch Foundation recorded 3,443 AI-generated child sexual abuse videos in 2025, up from just 13 the prior year.
[RESEARCH] MIT study: AI workers hit minimum benchmarks but struggle with quality across 11,000 tasks: MIT researchers tested 41 LLMs including Claude, Gemini, and ChatGPT across 11,000+ text-based job tasks, scored by humans with actual experience.

About 7min.ai · Subscribe to newsletter