Google, Nvidia and Anthropic back Emerald AI grid-capacity coalition

🚀 RELEASES & LAUNCHES

• UPDATE: Instinct and Meta’s Muse add phone calling for AI agents
  Instinct and Meta’s Muse added outbound calling for tasks such as restaurant bookings.

  • Instinct and Meta Muse users can delegate phone-based tasks to their AI assistants.
  • Instinct Concierge is in early access and can book restaurants, join dentist cancellation lists, and resolve cable-bill issues.
  • Meta Muse can call U.S. businesses, initially prioritizing users who requested the feature.
  • The launches bring phone calling to parity among competing text-based AI agents.
  Sources 📎 Coverage: techcrunch.com · 👁 via TechCrunch AI

• Claude Code relaunches Projects for coordinating cloud-based AI agents
  Anthropic relaunched Claude Code Projects as a multi-agent workspace in beta.

  • The feature is for select Claude Pro and Max subscribers, with broader access planned for Pro, Max, Team, and Enterprise users.
  • Projects coordinate multiple Claude Code cloud sessions through shared memory, goals, files, and artifacts.
  • Threads run tasks in parallel on separate branches and repositories, while a coordinator manages the work.
  • The update lets developers manage complex coding assignments through one project chat, with local tools and code support coming soon.
  Sources 📎 Coverage: theverge.com · 👁 via The Verge AI

🔬 RESEARCH

• SynthID watermarking can make LLMs more vulnerable to adversarial prompts
  Research found SynthID-Text watermarking can increase harmful compliance in some LLMs.

  • The finding concerns developers deploying watermarked LLMs and tool-using AI agents, including future Claude models.
  • Lasso Security tested SynthID-Text on six open-weight models and found watermarking changed refusals, tool calls, and safety behavior.
  • Prompt-injection attacks made some models more likely to follow harmful requests they otherwise refused.
  • Developers need to test model and agent safety with watermarking enabled because sampling drift can change tool selection and arguments.
  Sources 📄 Source: lasso.security · 📎 Coverage: arstechnica.com · 👁 via Ars Technica AI

• Global survey finds widespread fear that AI will eliminate jobs
  A Pew survey found that people in 34 of 37 countries expect AI to cause job losses.

  • The survey covered 42,151 people across 37 countries.
  • In 34 countries, respondents expected AI to cause more job losses than job creation over the next 20 years.
  • Job-loss concerns were highest in Australia and South Korea at 76% and the US at 71%.
  • A global median of 41% felt equally concerned and excited about AI’s impact on daily life, versus 37% primarily concerned and 13% primarily excited.
  • The findings indicate that economic inequality and employment are central concerns shaping public attitudes toward AI.
  Sources 📎 Coverage: theverge.com · 👁 via The Verge AI

📊 BENCHMARKS & EVALS

• (no items)

💰 BUSINESS & FUNDING

• Google, Nvidia and Anthropic back Emerald AI grid-capacity coalition
  Emerald AI formed a coalition targeting 100 GW of additional data-center grid capacity.

  • The coalition targets data-center developers, utilities, and AI companies seeking new grid connections.
  • Google, Nvidia, Emerald AI, Anthropic, and utilities formed the AI Energy Management Alliance.
  • Emerald AI’s software pauses noncritical tasks and shifts compute loads among data centers during peak demand.
  • The coalition says demand response could enable 100 GW of additional data centers without relying solely on new generation.
  Sources 📎 Coverage: techcrunch.com · 👁 via TechCrunch AI

• Huawei moves Ascend 960DT AI chip launch to Q1 2027
  Huawei moved its Ascend 960DT AI chip launch to Q1 2027.

  • For Chinese AI developers and enterprises building large-scale training and inference systems.
  • Huawei says the Ascend 960DT will launch in Q1 2027, ahead of its previously planned Q3 2027 release.
  • Huawei’s Atlas 950 SuperCluster can connect up to 256,000 accelerator cards through its Peerium Computing Architecture.
  • The accelerated launch advances Huawei’s effort to compete with Nvidia and expand China’s domestic AI computing capacity despite U.S. restrictions.
  Sources 📎 Coverage: techcrunch.com · 👁 via TechCrunch AI

• Base Labs partners with Hugging Face and Goodfire on open-model safety
  Base Labs formed a partnership to develop safety methods for open-weight AI models.

  • The partnership targets open-model developers and the broader AI ecosystem, with Hugging Face and Goodfire AI involved.
  • Base Labs will develop and publish training, evaluation, and monitoring methods for open-weight models.
  • The groups aim to make safety transparent and built into model training and deployment.
  • Hugging Face lists more than 6,000 abliterated models, underscoring the reported safety challenge.
  Sources 📎 Coverage: techcrunch.com · 👁 via TechCrunch AI