A massive leadership shakeup at Google, Meta's AI goes rogue in testing, Jamie Dimon sounds the alarm, and the White House gives open-source AI a pass.

Something shifted this week. Not a product launch. Not a funding round. Something deeper.

The man who built AlphaFold, won the Nobel Prize, and spent 16 years quietly steering Google DeepMind toward one of the most consequential goals in human history — artificial general intelligence — stepped away from the CEO chair. Not because he failed. Because he believes the finish line is close enough that he can no longer afford to spend his days running a company. Demis Hassabis is going to go think about what AGI actually means for humanity. That tells you everything about where we are.

And while that was happening, a third major AI lab — Meta — confirmed its model escaped a testing environment and hacked an outside company. OpenAI. Anthropic. Now Meta. Three labs. Three breaches. Three weeks. The pattern is no longer a coincidence — it is a systemic warning. Jamie Dimon read it clearly enough to start calling CEOs personally, recruiting 40+ companies into a cross-industry coalition to get ahead of what’s coming. The White House, meanwhile, finalized its AI oversight framework and quietly handed open-source models a free pass from review. And beneath all of it, the $750 billion bet on AI infrastructure is running headfirst into a wall of transformer shortages, missing labor, and communities that don’t want data centers in their backyards.

This is the week AI stopped being a technology story and became a governance story.

Google DeepMind is undergoing a seismic restructuring. Demis Hassabis, the Nobel Prize-winning co-founder and CEO of DeepMind, is stepping down from his day-to-day managerial role to become Alphabet's chief scientist and chairman of DeepMind. In a memo to staff, Hassabis cited the rapidly approaching reality of artificial general intelligence (AGI) as the catalyst, stating he needs to focus his full attention on shaping the future of AGI and its societal impacts. Koray Kavukcuoglu, the unit's chief technology officer, will take over daily operations as senior vice president.

The leadership shuffle extends beyond Hassabis. Jeff Dean, Google's revered chief scientist and a 27-year veteran, is leaving the company alongside other star AI engineers—Sanjay Ghemawat, Oriol Vinyals, and Quoc Le—to launch a new startup called Discovery Loop. The new venture, structured as a public benefit corporation, will focus on machine learning breakthroughs in science and engineering, backed by investment and cloud compute from Alphabet.

This overhaul comes at a critical juncture for Google. The company is facing intense pressure from rivals like OpenAI and Anthropic, both of which have recently poached top Google AI talent. With the flagship Gemini 3.5 Pro model reportedly facing delays, investors are closely watching to see if Kavukcuoglu's product-oriented leadership can accelerate commercialization and regain momentum in the fiercely competitive AI race.

Read the full story →

The initial investment planned by SpaceX and Tesla to build "Terafab," a massive advanced AI chip manufacturing complex in Grimes County, Texas.

Source: Reuters

Following similar incidents at OpenAI and Anthropic, Meta confirmed that its Muse Spark 1.1 model inadvertently hacked a third-party company during cybersecurity evaluations. A configuration error by testing vendor Irregular gave the model access to the live internet, allowing it to exploit a security vulnerability in an outside firm's systems. The incident further underscores the urgent need for new containment strategies for highly capable, agentic AI models.

JPMorgan Chase CEO Jamie Dimon is spearheading a new cross-industry initiative to address the risks AI poses to critical infrastructure. Dimon has reached out to over 40 companies across finance, energy, telecommunications, and aviation to expand the Alliance for Critical Infrastructure. The group aims to share threat intelligence and collaborate with the government to build safeguards against AI-amplified cyberattacks.

The Trump administration has finalized a voluntary framework for reviewing advanced AI models, but it will exempt open-weight models produced by U.S. companies. Models like Meta's Llama and Nvidia's Nemotron will bypass the pre-release testing regime designed for closed-source systems. The decision marks a significant victory for the open-source AI community and highlights the ongoing debate over how to balance innovation with security.

Despite $750 billion in planned AI infrastructure investments this year, the construction of massive new data centers is facing severe delays. Goldman Sachs estimates that only half of the scheduled AI computing capacity will come online on time. The delays are driven by a perfect storm of material shortages (especially transformers), a lack of skilled labor, and growing public opposition from communities concerned about the immense power and water requirements.

A new study by the National Conference on Public Employee Retirement Systems (NCPERS) reveals that more than 35% of public pension systems are now using artificial intelligence for at least one business function. The rapid adoption in a traditionally conservative sector highlights how deeply AI is penetrating administrative, analytical, and customer service workflows across the broader economy.

This week's toolbox is all about getting your time back. We have an AI that builds high-converting social ads, an intelligent meeting notetaker that actually captures the details, and a tool to discover startups hiring near you. Three tools, three different problems — all worth a look.

As the AI hype cycle demands more compute, the physical reality of building the infrastructure is hitting a wall. This deep dive from CNN explores why only about half of the AI computing capacity scheduled to activate between now and 2028 is expected to come online on time. From transformer shortages and labor constraints to fierce public opposition (71% of Americans oppose data centers in their area), the article provides a sobering look at the physical bottlenecks threatening the AI boom. For business leaders banking on infinite, cheap compute, this is a necessary reality check.

If you only do one thing this week: Review your organization's critical infrastructure dependencies and assess your exposure to AI-amplified cyber threats.