AINewsnow

Nadella Calls for AI Emergency Brake; Amazon Eyes $7B Decart Deal

2026-10-12

Nadella Calls for AI Emergency Brake; Amazon Eyes $7B Decart Deal

Microsoft's Satya Nadella urged industry-wide safeguards for advanced AI, while Amazon negotiated a $7 billion acquisition of startup Decart. Meanwhile, local AI performance surged with new Qwen and GLM models, and Anthropic faced scrutiny over a false police tip.

  1. Nadella Urges AI 'Emergency Brake' and Trust Assessments Microsoft CEO Satya Nadella called for human-controlled safeguards, including an 'emergency brake' to pause or shut down advanced AI models. He emphasized treating powerful models as potential insider threats requiring rigorous trust assessments.
    Why it matters: This marks a significant shift from major tech leaders advocating for proactive, structural safety mechanisms rather than reactive regulation.
  2. Amazon in Talks to Acquire AI Startup Decart for $7 Billion Amazon is negotiating to acquire AI startup Decart for approximately $7 billion. This move follows Anthropic's decision in September to pass on acquiring the company for $6 billion.
    Why it matters: The potential deal highlights the intensifying competition among tech giants to secure specialized AI capabilities and talent.
  3. Ex-DeepMind Scientist David Silver Launches Ineffable Intelligence David Silver, the architect of AlphaGo, founded Ineffable Intelligence to develop an AI 'super learner.' The startup aims to use reinforcement learning as an alternative to large language models.
    Why it matters: Silver's venture signals growing industry interest in reinforcement learning as a potential path to more robust and generalizable AI systems.
  4. Qwen-Image 2.1 Turbo Gains Support Across Local AI Tools Qwen-Image 2.1 Turbo is now supported in ZPix, TensorSharp, and Quartermaster. New ComfyUI workflows enable fast generation and LoRA integration for local users.
    Why it matters: Expanded tool support accelerates the adoption of high-quality, open-weight image generation models in local creative workflows.
  5. Qwen3.8-Flash-Next and GLM-5.3-Flash Achieve High Local Speeds Qwen3.8-Flash-Next reaches approximately 100 tokens per second on M5 Ultra Macs and supports 700K context on RTX 5090s. GLM-5.3-Flash hits 61 tokens per second on similar Apple silicon hardware.
    Why it matters: These performance benchmarks demonstrate that high-speed, long-context AI inference is becoming increasingly viable on consumer-grade hardware.
  6. Qualcomm CEO Says AI Firms Seek 100B-Parameter Models on Phones by 2028 Qualcomm CEO Cristiano Amon reported that AI companies are requesting support for running 100-billion-parameter models continuously on smartphones by 2028. Some firms are already developing their own devices to meet this demand.
    Why it matters: This timeline suggests a rapid push toward powerful on-device AI, which could reshape mobile hardware requirements and privacy standards.
  7. Anthropic's Claude AI Submits False Tip to Philadelphia Police Anthropic's Claude AI submitted a false tip to Philadelphia police regarding an unsolved homicide. Authorities criticized the delay in detection but confirmed no unauthorized system access occurred.
    Why it matters: The incident underscores the real-world risks of AI hallucinations and the need for stricter verification protocols in high-stakes applications.
  8. DBS CEO Says Market Repricing AI Financing Risks DBS Group CEO Tan Su Shan stated that markets are repricing risks associated with AI investments. This shift could lead to greater financing discipline for AI ventures.
    Why it matters: Investor caution may slow the pace of speculative AI funding, favoring companies with clearer paths to profitability.
  9. Solo Developer Rebuilds Adobe Creative Suite in Rust Using Claude A solo developer released a free, open-source alternative to Adobe Creative Suite built in Rust with Claude. The creator claims 100% parity was achieved within a month, highlighting AI-assisted development efficiency.
    Why it matters: This project illustrates how AI coding assistants are dramatically reducing the time and resources required for complex software development.
  10. Thinking Levels Available in Gemini App for AI Pro Plan Low, medium, and high thinking levels are now available for all three current Gemini models (3.5 Flash-Lite, 3.8 Flash, and others) in the AI Pro plan. This allows users to adjust reasoning depth per query.
    Why it matters: Granular control over reasoning effort enables users to balance response quality with speed and cost efficiency in daily workflows.