🤖 Microsoft Copilot ships Critique and Council multi-model modes 🤖
Multi-model integration arrives in Microsoft 365 Copilot Researcher with two new modes.

Microsoft rolled out two new multi-model features for Copilot Researcher in Microsoft 365. Critique uses a dual-model architecture where one model drafts research while Anthropic's Claude reviews it for source quality, completeness, and evidence grounding — outperforming single-model setups by 13.88% on the DRACO benchmark. Council runs both Anthropic and OpenAI models in parallel, surfacing agreements, disagreements, and unique insights from each. The updates ship alongside the broader rollout of Copilot Cowork into Frontier, Microsoft's Claude-based agentic tool for multi-step tasks. This marks a clear shift toward multi-model orchestration in enterprise AI, moving beyond single-model reliance to boost reliability.

🔥 xAI loses last cofounder besides Elon Musk 🔥
Nordeen's departure raises fresh questions about xAI's speed of leadership turnover.

Ross Nordeen, xAI's last remaining cofounder alongside Elon Musk, has left the company. His exit means Musk is now the sole original cofounder at the AI startup behind Grok. The departure follows a pattern of cofounder attrition at xAI, and leadership instability could directly impact recruiting, partnerships, and competitive positioning against OpenAI, Anthropic, and Google DeepMind. The news was widely covered across multiple outlets, reflecting just how closely the AI industry watches talent movement at major labs.

🧠 Meta's TRIBE v2 predicts brain responses to video 🧠
TRIBE v2 could replace traditional engagement metrics with neural prediction models.

Meta unveiled TRIBE v2, an AI model trained on fMRI brain scans that predicts how the human brain responds to video, audio, and text content. Instead of requiring physical brain scans, the model digitally simulates neural responses — going beyond likes and watch time to predict why content triggers specific brain reactions. Meta released the paper, code, and model under a non-commercial license while keeping commercial rights. The technology could serve as a blueprint for next-generation recommendation engines that optimize for predicted neural responses before content ever reaches users — a potential competitive edge in content distribution at scale.

INTERESTING TO KNOW

📊 Google opens TimesFM for time-series forecasting 📊

Google Research released TimesFM on GitHub, a rollout that brings pretrained foundation-model capabilities to time-series forecasting. Built on a patched-decoder attention architecture trained on a large temporal corpus, TimesFM generalizes across domains like finance, energy, healthcare, and supply chain — reducing the need for domain-specific model training. The approach mirrors how large language models scaled through pretraining, now applied to structured temporal data.

🔒 Anthropic ships Compliance API for Claude Platform 🔒

Anthropic launched a new Compliance API targeting regulation-heavy industries like finance, healthcare, and government. The API lets administrators audit logs, monitor user activities, and integrate compliance data into existing governance systems — tracking admin actions and resource-level events like file creation and deletion. As organizations deploy Claude in sensitive workflows, this tooling is critical for enterprise adoption in regulated sectors.