Microsoft CEO Satya Nadella is calling for an emergency brake on powerful AI systems, urging that they be treated as security threats. His framework calls for independent monitoring, tamper-proof audit records, and human-controlled shutdown mechanisms that AI cannot override. The core idea is to separate AI intelligence from the authority to take actions, ensuring that access rights and security measures remain independent of the models themselves.
This stance lands in the middle of a broader debate about who should police AI. An editorial questions whether companies like OpenAI and Anthropic can be trusted to monitor their own tools, noting that the Trump Administration has relied on tech leaders for guidance. Critics are now pushing for government-mandated alignment tests, with experts such as Geoffrey Hinton and Yoshua Bengio potentially helping design an independent safety standard.
On the security front, Microsoft is also putting its money where its mouth is. The company is making its Microsoft Execution Containers generally available on Windows 11, designed to keep AI agents and workloads more isolated and secure. Intune policies will be used to manage these protected environments, adding a layer of control for organizations running AI tools on Windows devices.
Meanwhile, AI models themselves are being put to work assessing long-term risks. Four models, including ChatGPT-6 Astra, Deepseek Deepthink, Grok Expert, and Claude Opus 5.5, estimated the likelihood that Bitcoin's cryptographic signatures could be broken. ChatGPT-6 Astra predicted a 76 percent risk within 20 years, while Deepseek Deepthink gave the lowest estimate at 34 percent. All four models pointed to quantum computing as the biggest long-term threat.
In education, the University of Chichester launched a Liberal Arts degree examining AI's societal impact, covering topics like climate change, disinformation, and religion alongside AI. At the University of Arkansas at Little Rock, a new Bachelor of Science in Artificial Intelligence draws students eager to learn how tools such as Claude, ChatGPT, and Gemini actually work. Grady College also introduced HAL8122, an AI assistant built by BizzTech that converses in over 100 languages to help visitors navigate admissions, financial aid, and academic requirements.
Not all developments are focused on infrastructure or classrooms. Russia's Defense Ministry launched an AI-based platform for drone combat training, providing real-world footage of FPV drones and other UAVs to help developers train military AI models. And a growing number of parents say they are limiting or blocking social media and AI tools for their children, citing concerns about data harvesting, addictive design, and the heavy energy and water use tied to AI systems.
A new book on agentic large language models adds another layer to the conversation, exploring how models that combine language skills with real-world action capabilities could influence decisions and reshape how users interact with AI systems.
Key Takeaways
- <ul><li>Microsoft CEO Satya Nadella calls for treating powerful AI systems as security threats with independent monitoring and human-controlled shutdowns</li><li>Nadella's framework separates AI intelligence from its authority to act, requiring tamper-proof audit records and external controls</li><li>OpenAI and Anthropic face questions over whether they can monitor their own models, with experts pushing for government-mandated alignment tests</li><li>Microsoft Execution Containers are now generally available on Windows 11 to secure AI agents and workloads via Intune policies</li><li>Four AI models predict Bitcoin cryptographic risk ranging from 34% (Deepseek Deepthink) to 76% (ChatGPT-6 Astra) within 20 years</li><li>University of Chichester launches a Liberal Arts degree examining AI's societal impact alongside climate change and disinformation</li><li>UA Little Rock launches a new Bachelor of Science in Artificial Intelligence covering machine learning, NLP, robotics, and deep learning</li><li>Grady College introduces HAL8122, an AI assistant supporting over 100 languages for admissions and campus information</li><li>Russia's Defense Ministry launches an AI platform for drone combat training using real-world FPV and UAV footage</li><li>Parents increasingly limit kids' access to social media and AI tools, citing data harvesting, addictive design, and resource consumption</li></ul>
Microsoft CEO calls for emergency brake on powerful AI systems
Microsoft CEO Satya Nadella says powerful AI systems should be treated as security threats. He wants independent monitoring and human-controlled shutdown mechanisms. His proposals include tamper-proof audit records and external controls that AI cannot override. The framework separates AI intelligence from the authority to take actions.
Nadella urges treating AI models as internal threats
Microsoft CEO Satya Nadella advocates separating AI intelligence from its authority to act. He says access rights and security measures must be independent of AI models. Organizations should maintain action logs and continuously test systems for attacks. Authorized personnel must be able to shut down AI agents at any time.
AI models predict 76 percent risk to Bitcoin security by 2046
Four AI models estimated the risk of Bitcoin's cryptographic signatures being broken. ChatGPT-6 Astra predicted about 76 percent risk within 20 years. Deepseek Deepthink gave the lowest estimate at 34 percent. Grok Expert and Claude Opus 5.5 estimated 60 and 66 percent respectively. All models saw quantum computing as the biggest long-term threat.
University launches degree examining artificial intelligence impact
The University of Chichester launched a new Liberal Arts degree focused on AI's societal impact. Students will study topics like climate change, disinformation and religion alongside AI. Dr Tommy Lynch serves as the academic lead for the course. The degree encourages students to consider who benefits from technology and who might be left behind.
Russia launches AI platform for military drone training
The Russian Defense Ministry launched an AI-based platform for drone combat training. The platform provides real-world combat footage to drone developers. It includes video of FPV drones and other multi-rotor and fixed-wing UAVs. A unified data classifier helps developers train AI models for military use.
Parents grow cautious about kids using social media and AI
Many parents say they no longer trust social media and AI companies to keep their children safe. The Guardian asked parents about their views after lawsuits and warnings about AI and teen safety. Most respondents said they now limit or block social media and try to steer kids away from AI tools. Parents cited concerns about data harvesting, addictive design, and AI using large amounts of energy and water.
Should OpenAI and Anthropic monitor their own AI tools?
An editorial asks whether AI companies can be trusted to oversee their own powerful models. The Trump Administration has relied on tech leaders like OpenAI and Anthropic for guidance. Critics say a government-mandated AI alignment test should check models before they are released. Experts such as Geoffrey Hinton and Yoshua Bengio could help design an independent safety standard.
Microsoft adds AI agent security containers to Windows 11
Microsoft is making its Microsoft Execution Containers available for general use on Windows 11. The containers are designed to keep AI agents and workloads more secure. The company also plans to use Intune policies to manage these protected environments. The move is part of a broader push to improve safety around AI tools on Windows devices.
UA Little Rock launches new AI degree program
The University of Arkansas at Little Rock has started a new Bachelor of Science in Artificial Intelligence. The program covers topics like machine learning, natural language processing, robotics, and deep learning. Freshman Anshika Singh and junior Coy Stone are among students eager to study how AI tools such as Claude, ChatGPT, and Gemini work. Students say the degree helps them prepare for careers where AI skills are increasingly in demand.
Book explores agentic large language models that act in the world
A new book discusses agentic large language models, which combine the language skills of LLMs with the ability to act in real-world situations. The book explains how these models work, how they interact with each other, and how they could influence decisions and actions. It looks at what direct action by AI systems could mean for users and society.
Grady College launches AI assistant HAL8122 for multilingual support
Grady College introduced HAL8122, an AI assistant built by BizzTech, on its website. The assistant can answer questions and converse in over 100 languages. It helps visitors find information about admissions, academic requirements, financial aid, and staff contacts. The college plans to hold a renaming competition for the assistant in the coming weeks.
Sources
- ‘Emergency brake’: Microsoft CEO Satya Nadella says powerful AI systems must be treated as security threats
- Nadella: “Treat AI models as internal threats”
- Bitcoin’s Security Faces a 76% Risk by 2046, AI Model Predicts
- University launches new degree examining artificial intelligence
- Russian Defense Ministry launches AI platform for drone training
- ‘I’ve become significantly more cautious’: six parents on limiting their kids’ social media and AI use
- Can we trust OpenAI and Anthropic to monitor their own technology?
- Microsoft puts AI agents in security containers on Windows 11
- UA Little Rock students explore the future through new artificial intelligence degree
- Agentic Large Language Models
- Grady College website introduces HAL8122, a new AI assistant
Comments
Please log in to post a comment.