A Tesla FSD crash with a sleeping driver highlights AI's most consequential gap: full-automation branding clashing with driver-responsibility reality. ⚖️
A quieter Saturday for AI news, but two threads stand out: Microsoft unveiled its 2-trillion-parameter MAI-Base-2 base model, raising the stakes in the open-weight race days after Meta's Muse Spark dropped, while CoreWeave's debt and equity selloff continues to test Wall Street's faith in the AI infrastructure buildout. Elsewhere, Reflection AI open-sourced its Beam model, Anthropic signed its first Canadian data-center lease, and India reported its first rogue-AI-agent incidents — a sign that agentic risk is going global fast.
Enterprise AI deployment stories today range from Itaú's full-scale generative AI rollout to Brazil's entire banking customer base, to a Tesla FSD crash that puts the limits of consumer-facing autonomous AI in sharp relief, to Waymo expanding its robotaxi footprint into Canada.
Developer tooling for multi-agent workflows edges forward with Band's new desktop app for orchestrating multiple coding agents simultaneously, while Raymond James shows how knowledge graphs are becoming critical middleware for grounding financial AI agents — a pattern increasingly generalizable across industries.
The open-weight model race accelerates on two fronts: Microsoft surfaces a 2-trillion-parameter base model positioned as more efficient than leading open-source alternatives, while Reflection AI throws its Beam weights and post-training recipes onto Hugging Face just a week after launch — lowering the barrier to customization for the broader developer community.
The AI infrastructure story today is bifurcated: on one side, meaningful new capacity moves (Anthropic's first Canadian data center, an offshore data-center energy experiment, optical-switching investment from AMD and Intel); on the other, CoreWeave's ongoing bond and equity rout keeps the market's anxiety about AI capex economics squarely in view.
Funding activity today is modest in scale but broad in ambition: RapidCanvas closes a $29M Series B for enterprise custom AI apps with backing from Vertex Ventures SEA & India and Accel, while Salience Labs' optical-switching raise (covered in Infrastructure) underscores that the AI supply chain continues to attract strategic capital from chip incumbents.
AI policy stress points are multiplying across geographies: Beijing is tightening export controls on AI technology amid US rivalry, Apple is updating App Store rules to explicitly ban AI-generated non-consensual intimate imagery after Grok-enabled deepfake scandals, and the US Director of National Intelligence is framing the OpenAI bugging arrest as a systemic warning about AI espionage risk to American leadership.
India registers two notable AI signals today: the country's first reported rogue-AI-agent incidents (credentials exposed in public repos and cloud misconfigurations) put it on the global agentic-risk map, while NITI Aayog is pushing MeitY to stand up a centralized data-center audit mechanism — a sign that AI infrastructure governance is becoming a domestic policy priority.
An Nvidia insider-turned-VC on why robotics hasn't hit its ChatGPT moment, yet — Amulya Vishwanath's 'Lego era' framing — that robotics still lacks standardized actuators, hands, software, and safety norms — is a useful corrective to the hype cycle and worth stress-testing against your own investment or product timelines. For anyone allocating resources to physical AI, her insider perspective on what's actually missing (and why the ChatGPT analogy breaks down) is the kind of calibration that saves expensive pivots. Read →