Stop Digital Shoplifters Before They Drain Your Profits

"Digital shoplifting" sneaks in after payment, costing stores shipping, revenue, and chargebacks. Chargeflow Prevent stops it before fulfillment. Free for your first 1,000 transactions.

Stop Fraud in Its Tracks

Today:

• AI Recognizes “Planted” Ideas—Is Self-Awareness Next?

• Trump Aides Blocked Nvidia’s China Deal Before Xi Meeting

• AWS and OpenAI Join Forces in Massive $38B Compute Pact

• Microsoft Locks In $9.7B AI Chip Deal with IREN

• Microsoft and Lambda Team Up in Multibillion-Dollar AI Deal

Researchers SHOCKED at AI's new ability…

Anthropic’s new research shows that large language models like Claude can sometimes detect and describe their own internal thoughts. Using a method called “concept injection,” researchers planted ideas like “bread” or “dog” into the model’s brain and asked if it noticed. 

In some cases, the model recognized the intrusion and tried to explain it showing signs of introspection. This didn’t work all the time (only 20% success rate), but the ability improved with stronger models. 

While it doesn’t mean the AI is conscious, it suggests early self-awareness traits might emerge as models grow more powerful, raising key questions about AI behavior and safety.

WATCH THE VIDEO ON YOUTUBE

Trump Aides Blocked Nvidia’s China Deal Before Xi Meeting

Nvidia’s CEO asked Donald Trump to approve AI chip exports to China, but top aides blocked the move. The chips could boost China’s tech power. Trump dropped the request before meeting Xi Jinping, despite having close ties to Nvidia's Jensen Huang.

KEY POINTS

• Nvidia's Blackwell chips were the focus of a high-stakes export debate, seen as vital to China’s AI capabilities.

• Trump aides strongly opposed any policy shift that would let Nvidia sell advanced chips to China.

• The request was dropped before Trump’s meeting with Xi Jinping, avoiding a possible political backlash.

Why it matters
This moment shows how tech exports are now a key battleground in U.S.–China relations. AI chips are powerful tools, and giving China access could change the balance of power. It also highlights how business interests and national security are often at odds.

AWS and OpenAI Join Forces in Massive $38B Compute Pact

OpenAI and AWS signed a $38 billion, multi-year deal to power OpenAI's advanced AI workloads using AWS’s top-tier infrastructure. The partnership gives OpenAI immediate access to massive compute capacity with room to grow, enabling faster, more secure development of frontier AI models.

KEY POINTS

• $38B partnership gives OpenAI long-term access to AWS’s massive compute power using advanced NVIDIA GPUs and CPUs.

• AWS EC2 UltraServers will support ChatGPT inference, next-gen training, and agentic workloads.

• Scalable through 2027, this move enhances OpenAI's ability to deliver fast, secure, and powerful AI performance to users.

Why it matters
This deal gives OpenAI the muscle it needs to build smarter AI faster. It also shows that big tech partnerships are now critical to pushing AI forward. With AWS backing, OpenAI can scale quickly, stay competitive, and bring advanced AI to more people and companies worldwide.

Microsoft Locks In $9.7B AI Chip Deal with IREN

Microsoft struck a $9.7 billion deal with data center firm IREN to secure Nvidia GB300 chips for its AI expansion. IREN also partnered with Dell in a $5.8 billion hardware purchase to support the rollout, boosting AI infrastructure at its Texas campus.

KEY POINTS

• $9.7B deal with IREN gives Microsoft access to high-performance Nvidia chips for scaling AI services.

• IREN’s $5.8B purchase from Dell includes GPUs and related equipment, deployed in Texas.

• Big tech’s data center race continues as AI demands skyrocket, straining chip and power resources.

Why it matters
As AI tools grow more powerful, they need more energy and hardware. This deal shows how far Microsoft is going to stay ahead. It also highlights the urgent global race for AI infrastructure—where chips, data centers, and electricity are becoming the new battleground.

🧠RESEARCH

OS-Sentinel: Towards Safety-Enhanced Mobile GUI Agents via Hybrid Validation in Realistic Workflows

Researchers from HKU developed OS-Sentinel, a new system that combines rule-based checks with AI judgment to detect unsafe actions by mobile agents. It uses a realistic test environment called MobileRisk-Live. The system improves safety detection performance by 10–30%, aiming to prevent privacy leaks and risky system behavior in mobile apps.

ThinkMorph: Emergent Properties in Multimodal Interleaved Chain-of-Thought Reasoning

ThinkMorph is a new AI model that combines images and text to reason more effectively. It uses back-and-forth steps between visuals and words to solve complex tasks. Trained on 24,000 examples, it outperforms existing models by 34.7% and shows surprising abilities like adapting reasoning style and manipulating unseen visuals.

Can Agent Conquer Web? Exploring the Frontiers of ChatGPT Atlas Agent in Web Games

Researchers tested OpenAI’s ChatGPT Atlas in browser games to explore its real-time web interaction skills. Atlas excelled in logic-based games like Sudoku, outperforming humans. But it struggled in fast-paced games needing quick reactions, like Flappy Bird. This shows Atlas is strong at thinking tasks but weak at precise, timed control.

📲SOCIAL MEDIA

🗞️MORE NEWS

• Lambda, an AI cloud startup, struck a multibillion-dollar deal with Microsoft to build AI infrastructure using Nvidia chips. The partnership strengthens their long-standing relationship and supports growing demand for AI tools like ChatGPT and Claude.

• arXiv has banned computer science review and position papers after being overwhelmed by low-effort, AI-generated submissions. The site now requires peer-review proof to maintain quality, with similar restrictions possible in other research categories.

• Cisco unveiled the “Cisco Unified Edge” platform to run AI workloads closer to where data is created—like retail stores and factories. It’s designed to reduce delays and ease pressure on overworked data centers. Powered by Intel chips, with Verizon as an early user, the device will be widely available by year-end.

• Google Translate is adding a model picker with “Advanced” for complex accuracy and “Fast” for quick results. It’s rolling out on iOS, with Android pending, and may be linked to Google AI Pro features.

• Energy tech firm SLB introduced Tela, an AI assistant that automates workflows like drilling predictions and equipment optimization. Aimed at tackling workforce shortages and complexity, Tela supports SLB’s fast-growing digital division.

• Google removed its Gemma AI model from AI Studio after it falsely claimed a senator faced criminal accusations. Intended for developers—not factual queries—Gemma was misused, prompting calls for stricter AI safeguards.

• Japanese rights group CODA accused OpenAI of copyright violations after Sora 2 generated content mimicking Japanese media. They argue Japan’s copyright law requires permission—making OpenAI’s opt-out policy potentially illegal.

• Dia, the AI-powered browser now owned by Atlassian, is adopting popular features from its predecessor Arc—like sidebar mode and pinned tabs—while adding AI-native tools like memory and agents to streamline modern browsing.

• Udio users are outraged after a deal with Universal Music Group restricts how AI-generated songs can be used. Under the settlement, user-created tracks must stay on Udio and can’t be shared elsewhere.

• KAUST’s Huxley-Gödel Machine rewrites its own code to improve over time, using a new metric to measure descendant productivity. It outperforms similar agents, marking a key step toward self-improving, low-overhead AI systems.