AI News: 18 Breaking Stories You Missed This Week

TL;DR

A weekly AI news roundup covering DeepSeek V4's disruptive pricing, open-weight model advances, legal and ethical drama, and new product launches.

Key Points

• 1.DeepSeek V4 is near state-of-the-art at a fraction of closed-model costs. At $1.74/million input tokens vs. GPT-5.5's $5 and Claude Opus 4.7's $5, it offers comparable benchmark performance with a 1M token context window, threatening US frontier labs.
• 2.Open-weight models are closing the gap with frontier models. Nvidia's Neotron 3 Nano Omni (vision, audio, language), Poolside AI's Laguna XS2 (33B params), and Mistral Medium 3.5 (128B dense) all launched as open weights this week, runnable locally for near-zero cost.
• 3.Anthropic faced major backlash over a billing bug that penalized users for mentioning rival tools. Users on $200/month Claude Max plans were charged extra or blocked when code referenced 'Hermes' or 'OpenClaw'; Anthropic initially refused refunds but reversed after posts hit 1.4M+ views.
• 4.Google signed a Pentagon deal allowing AI use for 'any lawful government purpose,' contradicting a 2014 promise to DeepMind founders that Google's AI would never be used for military or surveillance purposes; over 600 Google employees had previously signed a letter opposing the deal.
• 5.The Elon Musk vs. OpenAI trial began, with Musk performing poorly on the stand. Musk, who co-founded OpenAI, claims Altman and Brockman betrayed the nonprofit mission; after 5+ hours of testimony, he reportedly refused yes/no answers, argued with attorneys, and drew skeptical looks from jurors.
• 6.Microsoft and OpenAI restructured their partnership, removing the AGI clause and making Microsoft's license non-exclusive through 2032. The very next day, OpenAI announced its models, Codex, and managed agents would come to AWS, signaling a multi-cloud expansion strategy.
• 7.China blocked Meta's $2B acquisition of AI firm Manus despite the startup being Singapore-incorporated. The unwind is logistically complex — Manus staff have already moved into Meta offices, capital has been transferred, and executives have joined Meta's AI team.
• 8.Mayo Clinic's AI model can detect pancreatic cancer on routine CT scans up to 3 years before clinical diagnosis, identifying subtle disease signs before tumors are visible, back-tested successfully on historical scans of confirmed cancer patients.