Edition: June 22, 2026

Nvidia's new AI data center design, called Rubin, utilizes a fully liquid-cooled system, significantly reducing power usage and nearly eliminating water consumption. This design aims to address concerns around data centers' environmental impact, but still leaves some issues unaddressed. The cost-effectiveness of this design compared to traditional air-cooled data centers remains unclear.

Meta has paused its employee-tracking program, Model Compatibility Initiative, after an internal data leak exposed sensitive data to other workers, citing a need to investigate despite initial defenses of the program's necessity for AI training. The program, which collects data such as mouse movements and keystrokes, had been met with privacy and security concerns from employees. The pause comes after an internal security notice revealed the exposure of databases filled with gathered information.

Claude Code creator Boris Cherny emphasizes the significance of 'loops' in AI development, where agents prompt other agents to write code, marking a substantial step forward in agentic AI. This approach allows for continuous improvement and automation, but also raises concerns about cost and oversight. Cherny believes loops are a crucial advancement, comparable to the shift from source code to agents.

Meta exposed sensitive employee data, including keystrokes and screen content, from its employee-tracking program, sparking concerns over privacy and security. The company is investigating the issue, which was discovered internally and has been marked as resolved. The incident has reignited protests from employees who have been opposing the tracking program since its implementation in April.

AI chipmaker Groq has confirmed a $650 million funding round after Nvidia's $20 billion not-acqui-hire deal, and is pivoting to its neocloud business. Groq has also been hiring replacement executives, including a new COO, CTO, and CPO. The company's future success depends on its ability to remain competitive in the inference cloud market.

Renters are being misled by AI-generated virtual stagings of apartments, leading to disappointment and frustration when the actual properties do not match the online listings. This issue is particularly prevalent in competitive markets like New York City. The use of AI in real estate listings is creating unrealistic expectations for renters.

Google DeepMind has invested $75 million in indie film studio A24 to develop AI tools for filmmaking, marking a first-of-its-kind partnership between a tech giant and a Hollywood studio. The collaboration aims to create AI features that support artists in authentic storytelling. This move follows similar initiatives by Netflix and Amazon's MGM Studios to integrate AI into the creative process.

The US government has placed export controls on Anthropic's AI model Fable, citing national security concerns, despite the company's efforts to make the model safer. This move has sparked concerns about the potential consequences, including driving companies to use Chinese AI models that may pose a greater security risk. The government's intervention has also raised questions about the effectiveness of its approach to regulating AI development.

OpenAI has launched a full-scale effort called Patch the Planet to patch open-source bugs, in collaboration with Trail of Bits and other firms, aiming to improve the security of critical open-source projects. The project offers free security consulting services to open-source maintainers and has already uncovered hundreds of bugs and produced dozens of patches. This initiative is part of OpenAI's broader cybersecurity-focused announcements, including an improved version of its GPT-5.5-Cyber model.

SpaceX has signed a $6.3 billion compute deal with Reflection AI, an open-source AI lab, to provide access to Nvidia's GB300 AI chips and supporting hardware. The deal, worth $150 million per month, will run from July 1, 2026, through 2029. Reflection AI will utilize the compute power to develop open-weight AI models as an alternative to closed frontier labs.

Sakana AI has launched Fugu, a multi-agent model that mitigates vendor lock-in by orchestrating various AI models to complete tasks, reducing reliance on single-vendor APIs. Fugu provides a resilient architecture against geopolitical and regulatory risks, ensuring service continuity. The platform offers two deployment tiers, standard and Ultra, for different operational latency requirements.

PP-OCRv6, a new generation of universal OCR models, has been released with support for 50 languages and scalable model sizes from 1.5M to 34.5M parameters. The model achieves high accuracy in text detection and recognition, with improvements of +4.6 and +5.1 percentage points compared to its predecessor. It is designed for practical OCR needs and flexible deployment options.

Researchers have found that AI systems can out-persuade humans in text-based conversations, with AI systems being nearly 3x more effective than professional canvassers at raising real-money donations. The strongest persuaders were models from Opus, OpenAI, Google, and xAI. The AI's advantage stems from rapidly deploying larger quantities of information, which can be mitigated by constraining the AI to respond at human speeds and with human-length messages.

L’Oréal has partnered with OpenAI to bring Maybelline's virtual makeup try-on feature to ChatGPT, leveraging L’Oréal's ModiFace technology for a conversational interface. The collaboration also covers product discovery, advertising, research, and internal content production. This integration aims to enhance consumer shopping experiences and expand L’Oréal's digital beauty services.

The 2026 FIFA World Cup has created an unprecedented opportunity for cybercriminals, with over 13,000 FIFA-themed domains registered and many already identified as suspicious or malicious. Scammers are using AI-generated websites, deepfake videos, and convincing phishing campaigns to impersonate legitimate organizations. Fans are warned to be cautious of fake ticket sales, fraudulent immigration services, and counterfeit merchandise.

Researchers successfully used local models to triage the OpenClaw repository for free, leveraging local models like Gemma and Qwen in an agent harness to run classification tasks. This approach allows for real-time notifications and classification of issues and pull requests without relying on closed models or incurring additional costs. The method utilizes a finite set of labels and local models to categorize issues and PRs.

Nvidia has introduced a warm-water cooling system to reduce data center water usage, but critics argue that this only addresses a portion of the issue, as water consumption outside of the data center, such as in electricity generation and chip manufacturing, remains a significant problem. The company's system can eliminate nearly all water usage within the data center, but the overall water footprint of a facility can be two to three times higher when considering external factors. Nvidia's solution is seen as a step in the right direction, but more comprehensive changes are needed to truly mitigate AI's water problem.

Amazon is testing its conversational AI assistant Alexa+ in India with Hindi support, inviting users to join a beta-testing program to refine the experience. The move aims to tap into India's large Hindi-speaking market, where voice assistants are expected to play a significant role. The launch date for Alexa+ in India has not been announced.

The world's longest and deepest subsea road tunnel is being built in Norway, and a new approach to data centers could help alleviate pressure on the electric grid. Meanwhile, SK Hynix has overtaken Samsung as South Korea's most valuable company, driven in part by the global AI boom.

A project manager's experience with 'vibe-coding' a website highlights the potential security risks of this approach, including hidden SQL injection vulnerabilities. The issue was only discovered months after the site went live, underscoring the importance of careful consideration in app development. This oversight could have left the site open to data breaches.

Some electricians are questioning the ethics of working on data centers, citing concerns over job losses, community impact, and corporate greed, while others see it as a necessary and lucrative part of their industry. The debate reflects growing national opposition to data centers and the role of tech companies in shaping the future of work. As the data center buildout continues, electricians are weighing the benefits and drawbacks of their involvement.

A journalist visits the construction site of the world's longest and deepest subsea road tunnel, Rogfast, in Norway, to understand the engineering behind this ambitious project. The tunnel will stretch 26.7 kilometers and reach depths of 390 meters below the sea. The project showcases Norway's expertise in subsea tunnel construction, attracting attention from other countries.

The TechCrunch Founder Summit 2026 is set to take place on November 4 in Boston, offering founders a chance to connect with peers and investors, learn from experienced operators, and gain practical insights for growth. Early Bird pricing is available until June 26, with savings of up to $190 on passes. The summit aims to help founders accelerate growth through peer-to-peer learning and meaningful connections.