google, midjourney and gemini Updates

The Pentagon is rolling out a new AI platform, GenAI.mil, to 3 million employees, warfighters, and contractors this week. Defense Secretary Pete Hegseth emphasizes this move as the future of American warfare, with the department fully embracing artificial intelligence. Google's Gemini for Government powers the platform, enabling users to conduct deep research, format documents, and analyze videos or images quickly. The system is certified for Controlled Unclassified Information and Impact Level 5, ensuring secure operational use, and Google Cloud guarantees that Department of Defense data will not train Google's public AI models. While the military integrates advanced AI, businesses face challenges with "shadow AI," where employees use unapproved tools like ChatGPT, Midjourney, or Claude. This trend, driven by easy access and perceived productivity gains, creates significant risks including data breaches, compliance violations, and inconsistent work. Companies need clear policies and faster approval processes to manage these issues effectively and maintain trust. Despite the rapid adoption, IBM CEO Arvind Krishna expresses skepticism about short-term returns on massive AI investments, estimating an $8 trillion capacity investment would require $800 billion in profit just to cover interest. He also gives a low chance (0-1%) for current technology to achieve Artificial General Intelligence. Meanwhile, China leverages its manufacturing strength to lead in AI hardware, producing devices like AI glasses and credit card-sized meeting summarizers. This focus on practical, accessible AI helps integrate the technology into daily life. Looking ahead, Arm CEO Rene Haas forecasts that AI-powered humanoid robots will automate much of factory work within five to ten years, using "physical AI" for flexible reprogramming. Securing these advanced Edge AI devices, which run models directly on hardware for faster responses and better privacy, becomes crucial, with Arm building protections into its processor architectures. Furthermore, Mistral AI recently launched its Devstral 2 coding models, including a 123B version that is more cost-efficient than Claude Sonnet for real-world tasks, highlighting ongoing innovation in AI development. The concept of intelligence itself is also being re-evaluated, with past experiments like Toshiyuki Nakagaki's slime mold demonstrating problem-solving abilities without a brain, challenging traditional definitions.

Key Takeaways

• The Pentagon is deploying GenAI.mil, a new AI platform powered by Google's Gemini for Government, to 3 million employees, warfighters, and contractors for secure operational use.
• Google Cloud ensures that Department of Defense data used with Gemini on GenAI.mil will not train Google's public AI models.
• Businesses face significant risks from "shadow AI," where employees use unapproved tools like ChatGPT, Midjourney, or Claude, leading to potential data breaches and compliance issues.
• IBM CEO Arvind Krishna doubts short-term returns on current massive AI investments, citing high costs and a low probability (0-1%) of achieving Artificial General Intelligence with existing technology.
• China is gaining a major advantage in the global AI race by leveraging its manufacturing capabilities to produce a wide range of AI-powered hardware, including AI glasses and meeting summary devices.
• Mistral AI released Devstral 2, a new family of open-source coding models, with the 123B version proving more cost-efficient than Claude Sonnet for real-world tasks.
• Arm CEO Rene Haas predicts that AI-powered humanoid robots will automate large parts of factory work within the next five to ten years, using advanced "physical AI."
• Securing Edge AI devices is critical due to the potential for cyberattacks to cause real-world physical impacts, emphasizing the need for hardware trust and secure root of trust.
• Publishers are using AI citation tracking tools like Semrush and Similarweb to understand how AI systems, such as ChatGPT and Perplexity, utilize their content via Retrieval Augmented Generation.
• Research into organisms like slime mold challenges traditional definitions of intelligence, suggesting problem-solving abilities can exist without a brain or consciousness.

Pentagon gives 3 million staff new AI tools

The Pentagon is launching a new AI platform called GenAI.mil for 3 million employees, warfighters, and contractors this week. Pentagon Chief Technology Officer Emil Michael announced this major step, calling it a transformation for the department. Defense Secretary Pete Hegseth stated that the department is fully embracing artificial intelligence as a fighting force. The tools are certified for Controlled Unclassified Information and Impact Level 5, ensuring security for operational use. Google's technology will power features like natural language conversation and web-grounded search to provide reliable information.

Pentagon deploys Google Gemini AI for 3 million users

The Department of Defense will use Google's Gemini for Government in its new GenAI.mil platform. This allows 3 million Pentagon employees to use generative AI on their work computers. Defense Secretary Pete Hegseth said this marks the future of American warfare. Employees can use Gemini to do deep research, format documents, and analyze videos or images quickly. Google Cloud provides the technology, ensuring that DOD data will not train Google's public AI models.

US military launches GenAI.mil with Google Gemini

The US Department of War launched its GenAI.mil platform to bring advanced artificial intelligence to military operations. Secretary of War Pete Hegseth announced that Google Gemini is the first AI model integrated into the system. This platform will help soldiers conduct deep research, format documents, and analyze video or images very quickly. Hegseth emphasized that AI is the future of American warfare and that all technology used will be American made. The GenAI.mil website faced some technical issues right after the announcement.

Understanding and managing shadow AI risks

Shadow AI refers to employees using AI tools like ChatGPT or Midjourney without company approval or oversight. This trend is growing because these tools are easy to access and seem to boost productivity. However, shadow AI creates serious risks for organizations. These risks include data breaches, breaking compliance rules, and producing inconsistent work. It also makes it hard to track decisions and can damage trust. Companies need clear rules and faster approval processes to manage shadow AI effectively.

Businesses face risks from unapproved AI tools

Shadow AI happens when employees use AI tools like ChatGPT or Claude without their company's official approval. This often occurs because these tools are easily available and seem to make work faster. However, using unapproved AI can lead to big problems such as data breaches and breaking important privacy rules. It also causes issues with inconsistent work, makes auditing difficult, and can harm customer trust. Companies should create clear policies and improve approval processes to handle shadow AI.

IBM CEO doubts short-term AI investment returns

IBM CEO Arvind Krishna believes that current massive investments in artificial intelligence will not yield short-term returns. He estimates that an $8 trillion investment for AI capacity would require $800 billion in profit just to cover interest. Krishna also thinks that current technology is unlikely to lead to Artificial General Intelligence, giving it a 0-1% chance. He points out that AI systems depreciate quickly and need retooling within five years. Despite his concerns about investment returns, Krishna still sees AI unlocking trillions of dollars in enterprise productivity.

Publishers track AI citations for content use

AI citation tracking helps publishers understand when and how their content is used by AI tools like ChatGPT and Perplexity. These AI systems use a process called Retrieval Augmented Generation to summarize information and often show citations as clickable links. This tracking is important because the way people search for information has changed, with more users turning to AI. Unlike traditional search, AI responds to open-ended questions, making it harder for publishers to know why their content is cited. Tools like Semrush and Similarweb help automate this process, but results can vary.

Securing advanced Edge AI devices

Edge AI, which runs AI models directly on devices, is growing fast due to benefits like faster responses and better privacy. However, this also creates new security risks, as cyberattacks can now have real-world physical impacts. A malicious update, for example, could shut down factory production. To protect these systems, experts emphasize hardware trust, starting with a secure root of trust for every edge device. Arm, a major chip designer, has developed schemes like PSA Certified and built protections directly into its processor architectures to make edge AI more secure.

Slime mold shows new ways to define intelligence

In 1995, biologist Toshiyuki Nakagaki showed that slime mold, a brainless organism, could solve complex problems. He recreated the Tokyo rail system with oat flakes on a gel plate, and the slime mold rebuilt the network in just 24 hours. This experiment challenges the common belief that intelligence requires a brain or consciousness. It suggests we should broaden our understanding of what "smart" means. Today, many focus on Artificial General Intelligence that thinks like humans, but Nakagaki's work points to other forms of intelligence.

Google reviews its 2025 highlights

Google published a quick review of its most exciting launches and biggest moments from 2025. The company shared this update as it prepares for the new year. This recap helps users remember key developments and innovations from Google throughout the past year.

Mistral AI launches Devstral 2 coding models

Mistral AI released Devstral 2, a new family of coding models, along with the Mistral Vibe CLI. Devstral 2 comes in two sizes, 123B and 24B, both open-source and designed for distributed intelligence. The 123B model achieved 72.2% on SWE-bench Verified and is much more cost-efficient than Claude Sonnet for real-world tasks. Devstral Small 2, the 24B model, can run on consumer hardware and supports multimodal agents. Mistral Vibe CLI is an open-source command-line assistant that helps with code automation using natural language.

China leads AI hardware with manufacturing strength

China is gaining a major advantage in the global AI race by using its strong manufacturing abilities to create many AI-powered devices. Experts like Dr. Kai-Fu Lee believe China's skill in large-scale production will drive the next wave of AI adoption through physical products. The market already offers over 70 Chinese companies making AI glasses with features like real-time translation. Other innovations include the Alibaba DingTalk A1, a credit card-sized AI device for meeting summaries, and neck-worn gadgets for language learning. This focus on practical, accessible AI solutions helps integrate AI into daily life and work.

Arm CEO predicts robots will automate factories

Arm CEO Rene Haas predicts that AI-powered humanoid robots will automate large parts of factory work within the next five to ten years. He explains that these general-purpose robots, with advanced "physical AI," can be easily reprogrammed for different tasks, unlike older factory machines. Haas believes this shift could help balance global manufacturing competition. He also mentioned that the semiconductor supply chain has many single points of failure, like TSMC and ASML, which was evident during the COVID-19 chip shortages.

Sources

• DOD initiates large-scale rollout of commercial AI models and emerging agentic tools
• U.S. military to use Google Gemini for new AI platform
• US Dept Of War Creates GenAI.Mil Platform That Makes AI Models, Starting With Google Gemini, Available To Soldiers
• What is shadow AI and what can you do about it?
• What is shadow AI and what can you do about it?
• Arvind Krishna, IBM CEO, on AI investment: “It’s my view that there’s no way you’re going to get a return on that”
• WTF is AI citation tracking?
• Edge AI is advancing. So are the ways in which it's being protected
• How slime and dumb rocks can help us better define “smart”
• 2025 at Google
• Introducing: Devstral 2 and Mistral Vibe CLI.
• China’s AI Hardware Surge: A Manufacturing-Driven Edge in the Global Race
• Physical AI will automate ‘large sections’ of factory work in the next decade, Arm CEO says