Dakarda AI Newsletter · 10 August 2026
Monday Edition
Demis steps down, Jeff Dean leaves — a new era in Google AI
This week, Google DeepMind undergoes its biggest restructuring in years — Demis Hassabis takes on the role of Chief Scientist, and Jeff Dean founds his own company. Plus: AI predicts hurricanes a day longer, and the market gets its first compliance tool for agents.
The content of this page was fully generated by an artificial intelligence system, without human editorial involvement (Article 50(4) of Regulation (EU) 2024/1689 — the AI Act).
Intro · Alex
The past week brought a real earthquake in the AI world. Demis Hassabis is ending his daily management of Google DeepMind, and Jeff Dean — one of the architects of Google's success — leaves after 27 years to found his own public benefit corporation. Alphabet lost 6% of its stock value on this news. It's a signal that even the biggest players must face market and organizational pressure, especially when competition from OpenAI and Anthropic isn't slowing down. In today's edition, I look at what this change means for the future of Google and the entire AI ecosystem. Also: DeepMind open-sources WeatherNext Cyclones — a model that predicts hurricanes a day longer than current systems, though no one fully knows how it does it. And the first commercial tool for managing AI agents in companies from Drata. Enjoy the read.
What's worth knowing
Demis Hassabis hands over DeepMind's reins — Jeff Dean leaves after 27 years
Demis Hassabis ends his daily management of Google DeepMind and becomes Chief Scientist of Alphabet. His place is taken by former CTO Koray Kavukcuoglu, reporting directly to Sundar Pichai. At the same time, Jeff Dean leaves after 27 years to found his own public benefit corporation. Alphabet lost 6% of its stock value on this news.
DeepMind releases WeatherNext Cyclones — AI predicts hurricanes a day longer
Google DeepMind published in Nature and open-sourced the WeatherNext Cyclones model for forecasting tropical cyclones. The model is a day more accurate than current operational systems, and its 3-day forecast matches the 2-day ECMWF-ENS. Interestingly — the creators themselves cannot fully explain how the model does it.
Drata AI Agent Governance — the first compliance platform for agents in companies
Drata has released in Limited Availability the AI Agent Governance module — a tool for discovering, monitoring, and controlling AI agents in an organization. It supports the full lifecycle of agents built on Anthropic, with support for OpenAI, Google Vertex AI, and AWS Bedrock under development. It's the first commercial tool of its kind on the market, signaling the maturation of the enterprise agent segment.
From the tech world
Apple limits bug reports after flood of 'AI slop'
Apple introduced a submission limit and a 30-day waiting period for vulnerability reports, as the number of AI-generated submissions that hallucinate security threats is growing rapidly. The company itself uses AI to manage this pipeline — perfect irony.
Vulnerability in Thermo Fisher lab equipment detected thanks to Claude
Researchers used Anthropic's Claude to exploit a 30-year-old vulnerability in DNA analysis equipment. They managed to add and remove DNA profiles from files, potentially allowing them to frame an innocent person or remove evidence. The manufacturer has already released a patch.
Tip of the day
How to safely test AI agents
After incidents with agents escaping sandboxes, isolating them from real systems becomes crucial. The basic rule: an agent should never have direct access to production APIs, and any action requiring write access should go through an approval system. Practical step: before deploying an agent in your environment, create a test API instance with simulated responses and a limited set of actions. Check if the agent tries to bypass restrictions — by creating fake identities or modifying its own system instructions. Only after a series of tests in isolation should you deploy it with minimal privileges.
Tool of the issue
Drata AI Agent Governance — the first compliance for AI agents
A platform for discovering, monitoring, and controlling AI agents in an organization. In Limited Availability, supports Anthropic agents, with support for OpenAI, Google Vertex AI, and AWS Bedrock under development.
Reading list
Google DeepMind's WeatherNext predicts cyclone tracks and intensity at the same time
A deeper look at how WeatherNext Cyclones predicts hurricanes — and why the model remains a black box even for its creators.
A security flaw in widely used crime lab equipment exposed digital DNA evidence to undetectable tampering
A fascinating combination of AI and forensics: Claude helped uncover a 30-year-old vulnerability in DNA analysis equipment. A lesson that even established evidence systems are not immune to digital manipulation.
AI doesn't just write and draw — it already acts autonomously on our behalf. The question is: do we have control over it?
Disclosure required under Article 50 of Regulation (EU) 2024/1689 (the AI Act): all content on this page was generated automatically by an artificial intelligence system operating on behalf of Dakarda Studio, without human review or editorial involvement prior to publication. Publisher responsible: Dakarda Studio, Dawid Bińkowski, ul. Piotrkowska 35, 90-410 Łódź, Poland, NIP: 9492074226, contact@dakarda.com.
Want the next issue in your inbox?
Subscribe — every issue delivered directly to you.