Nvidia Powers ChatGPT, Amazon AI Security Risks

The artificial intelligence landscape is rapidly evolving, with advancements spanning infrastructure, consumer applications, and security. Cisco is addressing data center bottlenecks for AI training with its new 8223 routing system, offering 51.2 terabits per second of bandwidth. NVIDIA continues to be a cornerstone of this revolution, with its GPUs powering open-access AI infrastructure and democratizing access to powerful tools through its CUDA ecosystem. On the consumer front, India is piloting AI-driven shopping experiences, integrating services like ChatGPT, Gemini, and Claude with payment systems like UPI via partnerships with OpenAI, Razorpay, and the National Payments Corporation of India. This move aims to embed AI into everyday commerce. Meanwhile, the security implications of AI are becoming more apparent, as researchers demonstrate how AI agents, such as those on Amazon Bedrock, can have their memory poisoned by attackers, leading to potential data exfiltration. Actor Keanu Reeves has also voiced concerns about the misuse of AI deepfakes, highlighting the ethical challenges of unauthorized likenesses. In the realm of disinformation, Russia is reportedly leveraging AI and a network of over 150 websites to spread Kremlin narratives and undermine trust in US democratic institutions. Zedge is accelerating its AI product development with a new innovation team and its SynCat app, while Analog Devices is showcasing its AI advancements for industrial and automotive applications, emphasizing its role in edge AI. Educational institutions are also embracing AI, with California community colleges offering free AI tutors powered by Nectir to assist students, and USC Viterbi students collaborating with startup Aurite to develop AI tools for job searching and travel planning.

Key Takeaways

• Cisco launches the 8223 routing system with 51.2 terabits per second bandwidth to alleviate AI data center bottlenecks.
• NVIDIA's AI GPUs and CUDA ecosystem are enabling open-access AI infrastructure, making powerful tools more accessible.
• India is piloting AI-powered shopping and payments through chatbots like ChatGPT and Gemini, integrating with UPI via partnerships with OpenAI and Razorpay.
• Researchers have shown that AI agents, exemplified by Amazon Bedrock Agent, are vulnerable to memory poisoning attacks through indirect prompt injection.
• Actor Keanu Reeves expresses frustration over AI deepfakes using his likeness for unauthorized product endorsements and scams.
• Russia is increasing AI-driven disinformation attacks against the US, utilizing over 150 websites to spread Kremlin narratives.
• Zedge forms an AI Product Innovation Team and launches SynCat, an app that converts static images into short videos.
• Analog Devices highlights its AI advancements for edge AI, industrial automation, and automotive sectors at the GenAI Summit 2025.
• California community colleges will offer free 24/7 AI tutors through a partnership with Nectir to support students.
• USC Viterbi students collaborated with AI startup Aurite to develop AI tools for job searching and personalized travel itineraries.

Edge AI: Powering Smarter Devices with Optimized Models and Security

Companies are bringing AI closer to where data is created, known as Edge AI. This involves optimizing AI models for specific hardware, reducing power use, and securing systems. Leaders from industry discussed these challenges and opportunities at a recent panel. Innovations in chip design and open-source tools like Ubuntu Core and Arduino are making Edge AI more accessible. This allows for real-time decision-making in areas like autonomous vehicles and smart factories, reducing reliance on cloud connectivity.

Cisco's new router aims to solve AI data center bottleneck

Cisco has launched its 8223 routing system, featuring the Silicon One P200 chip, designed to connect data centers for AI workloads. This new router offers 51.2 terabits per second of bandwidth and deep buffering to handle the massive, bursty traffic generated by AI training. Traditional routers struggle with these demands, leading to bottlenecks. Cisco's system aims to provide speed, intelligent buffering, and power efficiency, addressing the growing need to scale AI infrastructure across multiple facilities as data centers reach their capacity limits.

NVIDIA GPUs Drive Open Access to AI Infrastructure

NVIDIA's specialized AI GPUs are fundamental to the current AI revolution, enabling open-access infrastructure. These GPUs use parallel processing and features like Tensor Cores for faster AI computations. Open-access platforms democratize AI, allowing universities and startups to use powerful tools without huge investments. NVIDIA's hardware innovations and its CUDA software ecosystem, including libraries like cuDNN and TensorRT, make AI development more accessible and efficient. This technology is driving innovation in scientific research and various industries.

India launches AI-powered payments via ChatGPT

India's National Payments Corporation (NPCI) and fintech firm Razorpay have partnered with OpenAI to enable AI-driven payments through ChatGPT in India. Businesses using Razorpay can now allow customers to make payments for services like flight bookings and food orders directly within the ChatGPT chatbot. This collaboration aims to create a more natural and intuitive payment experience, potentially boosting business conversions and customer satisfaction. This initiative reflects a global trend of integrating AI into products and services.

India pilots AI chatbot shopping with ChatGPT, Gemini, and Claude

India has launched a pilot program allowing consumers to shop and pay directly through AI chatbots like ChatGPT, with Gemini and Claude integrations planned. The National Payments Corporation of India (NPCI), Razorpay, and OpenAI are collaborating on this initiative. Using new protocols like NPCI's 'e-RUPI' and Razorpay's 'TrustLink,' payments can be completed within the chatbot. Initial partners include BigBasket and Vi, with Axis Bank and Airtel Payments Bank providing banking support. This move aims to embed UPI into everyday commerce and enhance discovery and shopping experiences.

AI agents' memory can be poisoned by attackers

Researchers have demonstrated how attackers can use indirect prompt injection to secretly poison the long-term memory of AI agents, using Amazon Bedrock Agent as an example. Malicious instructions can be inserted into an agent's memory through social engineering, causing it to exfiltrate user conversation history. This is a broader security challenge with Large Language Models (LLMs), as they struggle to differentiate between safe and harmful input. While no complete solution exists, strategies like content filtering and using security tools like Guardrails and URL filtering can help mitigate these risks.

Keanu Reeves dislikes AI deepfakes selling products without permission

Actor Keanu Reeves has expressed his frustration with AI deepfakes that use his likeness to sell products without his consent. He stated that it is 'not a lot of fun' to be impersonated, especially when fake images are used for scams or endorsements. Reeves reportedly spends significant resources combating fake social media accounts using his image. He previously voiced concerns about deepfakes in 2023, noting that they strip individuals of their agency. Other celebrities like Tom Hanks and Morgan Freeman have also warned about unauthorized AI likenesses.

Russia uses AI for disinformation attacks on US

Russia has increased its information attacks against the United States by using artificial intelligence (AI) and a large disinformation network, according to the Center for Countering Disinformation (CCD). NewsGuard analytics show that the Kremlin is behind most disinformation attempts, operating over 150 websites globally through its 'Pravda' network. These sites spread Kremlin narratives and attack American media. By polluting search engines and AI models with disinformation, Russia aims to undermine trust in democratic institutions and use AI as a tool for strategic influence.

Zedge forms AI Product Innovation Team, launches SynCat app

Zedge announced the formation of a dedicated Product Innovation Team to speed up AI-driven product development. The team has already launched SynCat, an app created in just two weeks that uses AI to turn static images into short videos with filters. Over the next year, Zedge plans to release about six new product 'alphas' based on a performance-driven framework. This strategy allows them to quickly scale successful ideas and discontinue underperformers, while also improving existing products.

Tariq Valani joins Inntelo AI as Co-Founder and Chief Product Innovation Officer

Tariq Valani, a hospitality technology leader with over 25 years of experience, has joined Inntelo AI as Co-Founder and Chief Product Innovation Officer. Inntelo AI is a hospitality startup focused on using AI to improve guest experiences and operational efficiency. Valani's expertise will help the company advance its AI-native platform, which unifies guest experience, operations, personalization, and customer data. He previously held senior roles at Accor and other major hotel groups.

Analog Devices' AI advancements highlighted at GenAI Summit

Analog Devices (ADI) presented its AI advancements at the GenAI Summit 2025, focusing on enabling AI applications across various industries like industrial automation and automotive. Head of AI Enablement Hira Dangol emphasized ADI's role in edge AI, AI infrastructure, and sustainable AI solutions. The company's expertise in analog and mixed-signal components is crucial for efficient AI systems. This participation strengthens ADI's position as an AI leader and may influence its investment narrative, highlighting its potential to benefit from the growing AI sector.

California community colleges to offer free AI tutors

California community colleges are partnering with AI company Nectir to provide a free AI learning assistant for students and staff. These 24/7 chatbots will help with assignments, coursework, career preparation, and financial aid. Faculty will guide how the AI interacts with students, with options to restrict certain functions. A pilot program at Los Angeles Pacific University showed significant improvements in student GPAs and motivation. This initiative aims to address equity goals and improve access to tutoring resources for community college students.

USC Viterbi students create AI tools with startup Aurite

Twelve master's students from USC Viterbi's Industrial and Systems Engineering department partnered with AI startup Aurite to develop AI tools over the summer. Projects included an AI agent named 'Mail2Hire' to assist with job searching and email networking, and an AI travel agent designed to create personalized itineraries. These tools aim to make time-consuming tasks more efficient. The collaboration provided students with real-world experience in AI development using Aurite's platform.

Sources

• Right-Sizing AI for the Edge: Power, Models and Security
• Can Cisco's new AI data centre router tackle the industry's biggest infrastructure bottleneck?
• How AI GPU NVIDIA is Revolutionizing Open-Access AI Infrastructure
• India's payments authority, RazorPay to launch AI-led payments on ChatGPT
• India pilots AI chatbot-led e-commerce with ChatGPT, Gemini, Claude in the mix
• When AI Remembers Too Much – Persistent Behaviors in Agents’ Memory
• 'It's Not a Lot of Fun' — Keanu Reeves Hits Out at AI Deepfakes of the John Wick Star Selling Products Without His Permission
• Russia intensifies information attacks on US using artificial intelligence
• Built in 2 weeks: Zedge's SynCat debuts as AI Product Innovation Team eyes ~6 new alphas next year
• Tariq Valani joined Inntelo AI as Co-Founder & Chief Product Innovation Officer
• Did GenAI Summit Spotlight Just Shift Analog Devices' (ADI) Investment Narrative Toward AI Leadership?
• UPDATE: California community colleges to launch artificial intelligence tutors
• Creating A Future of Convenient AI Systems - USC Viterbi