AI Agent Shopping Surges 38-Fold, While AI Coding Tools Are Being Used to Remove Advertisements from Smart TVs

Nazrana Times | Special Report | October 21, 2026

Artificial Intelligence (AI) is rapidly transforming technology, business, cybersecurity and the protection of personal information worldwide. While these developments are creating new opportunities, they are also raising serious concerns about digital security, privacy, transparency and accountability.

Five major stories highlighted in the weekly AI newsletter Maze of Bot demonstrate that artificial intelligence is no longer limited to answering questions or generating content. AI agents are increasingly being used for cybersecurity operations, online shopping, software development and managing digital devices.

AI Agents Used to Breach More Than 100 Companies

According to a Reuters exclusive report cited in the newsletter, a Chinese-speaking hacker allegedly used commercially available AI agents to target more than 1,200 websites and successfully gain unauthorized access to over 100 organizations.

The affected organizations reportedly included major companies such as Deere, SharkNinja and Cummins.

One of the most striking aspects of the operation was the simplicity of the instructions given to the AI agents. The hacker reportedly used commands containing fewer than 100 Chinese characters, allowing the agents to perform much of the technical work independently.

According to the report, each successful intrusion took approximately 23 hours on average, while the entire operation cost around $5,000.

This translates to an estimated cost of just $50 per successful breach.

The companies involved reportedly described the damage as limited, indicating that the attacks primarily affected secondary websites, vendor-operated pages and supporting online services rather than their main systems.

Nevertheless, the incident demonstrates how AI technology could significantly increase the capabilities of cybercriminals.

Cybersecurity experts have repeatedly emphasized the importance of protecting not only primary websites but also older digital infrastructure, support portals, third-party platforms and forgotten online services.

The growing accessibility of AI-powered tools could make cybersecurity threats more widespread and difficult to detect.

Hidden AI Instructions Discovered on WhatsApp Help Pages

Another significant development involves WhatsApp’s customer support system and its use of artificial intelligence.

According to the newsletter, researchers at Redact identified 102 blocks of hidden instructions across 80 WhatsApp help pages.

These instructions were reportedly invisible to ordinary users viewing the pages but could be read by AI systems processing their content.

One particularly concerning example appeared on a support page intended for users who had lost their phones and were experiencing difficulties deleting their WhatsApp accounts.

The hidden instructions reportedly told AI systems not to recommend contacting a data protection regulator.

Another instruction discouraged references to privacy legislation, including the European Union’s General Data Protection Regulation (GDPR), even when users raised such issues themselves.

WhatsApp explained that the instructions had been created for its support chatbot and had appeared on the help pages because of a technical bug.

The company stated that the issue had subsequently been fixed.

However, researchers have not established whether the chatbot actually followed these instructions or withheld information about users’ legal rights.

The discovery raises important questions about transparency in AI-powered customer support.

If companies provide hidden instructions to their chatbots, users may not always know whether the information they receive is complete or influenced by internal corporate policies.

The incident also highlights the importance of independent oversight and ensuring that consumers have access to accurate information about their legal rights.

Meta Faces Privacy Questions After More Than 6,800 AI Chat Links Were Found Online

A third report highlights potential privacy risks associated with Meta’s AI agent application, Muse.

According to cybersecurity firm Claroc, more than 6,800 links to shared AI conversations were discovered in publicly accessible internet records.

The investigation reportedly found that the web addresses of shared conversations included their titles.

As a result, the URLs themselves could reveal information about the subjects discussed, even before someone opened the corresponding pages.

Claroc claimed that some accessible conversations contained potentially sensitive business and personal information.

Examples reportedly included a company’s revenue review, a small-business loan application and an annual employee calendar.

Following the findings, Meta temporarily suspended link-sharing functionality in Muse.

The company explained that shared pages were intended to be accessible only to people possessing the relevant links.

However, Meta did not confirm the reported figure of more than 6,800 shared conversations.

The incident has renewed concerns about the handling of sensitive information within AI-powered services.

As AI agents become increasingly integrated with emails, calendars, financial documents and business files, sharing a conversation could potentially expose more information than users originally intended.

The case highlights the need for stronger privacy controls, clearer sharing permissions and greater awareness among users before publishing AI-generated conversations or documents.

Online Shopping Through AI Agents Increases 38-Fold

Artificial intelligence is also beginning to play a more active role in online commerce.

Instead of merely helping consumers search for products or compare prices, AI agents are increasingly participating in the purchasing process.

According to figures Stripe shared with Upstarts Media, purchases made by AI agents through its Link checkout wallet increased 38-fold during September.

The total dollar value of these transactions reportedly increased 20-fold during the same period.

Retail purchases accounted for a significant share of the activity, with an average transaction value of approximately $100.

The number of shopping agents using Link reportedly increased from around 70 to approximately 200.

However, Stripe did not disclose the original number of transactions used to calculate the reported growth.

Therefore, the 38-fold increase should not be interpreted as evidence that AI agents now account for a large proportion of overall e-commerce activity.

Part of the increase was also associated with the launch of Meta’s Muse platform in early September.

Despite these limitations, the figures indicate an emerging shift in digital commerce.

AI agents may increasingly assist customers by identifying suitable products, comparing prices, evaluating options and completing authorized transactions.

For online retailers, this development could create new opportunities and challenges.

Businesses may need to ensure that their product information, pricing and checkout systems are accessible and understandable to AI-powered shopping assistants.

At the same time, consumer authorization, payment security and transaction transparency will remain essential.

Claude Code Used to Develop Software That Removes Smart TV Advertisements

Another development demonstrates how AI-powered coding tools are enabling individuals to modify and improve their digital experiences.

An open-source project called Free My TV has been developed to provide an alternative home-screen experience for Google TV devices.

The software is designed to remove sponsored content rows, advertising banners and automatically playing promotional material.

Instead, it focuses on displaying the applications that users actually want to access.

According to the newsletter, the project’s creator used Anthropic’s AI coding tool, Claude Code, during development.

The project reportedly gained approximately 1,500 GitHub stars within about one day.

It also includes instructions intended to help AI coding assistants adapt the software for other television platforms, including LG and Samsung devices.

However, installing or modifying this type of software may require developer mode and advanced system access.

Such modifications can involve compatibility, stability and security risks and should not be treated as ordinary app installations.

The project nevertheless illustrates how AI-assisted programming is making software development more accessible.

People with limited traditional programming experience may increasingly be able to create customized tools and alternatives to commercial software.

This trend could also encourage technology manufacturers to reconsider the amount of advertising and promotional content included in their products.

Four Emerging AI Tools Highlighted in the Newsletter

In addition to the five major stories, the newsletter introduced four AI tools attracting attention for their capabilities in video generation, text processing, web search and image creation.

Kling 4.0: AI Video Generation

Kling 4.0 was presented as an AI video-generation tool capable of producing clips lasting up to 30 seconds with stereo audio.

The newsletter also highlighted a faster, lower-cost 720p option.

Such technology could be useful for digital content creators, news organizations, marketing professionals and video producers.

Mercury 3: Faster AI Text Generation

Mercury 3 was described as an AI model designed to generate multiple words simultaneously rather than producing text strictly one token at a time.

According to the newsletter, its reported input pricing was $0.20 per million tokens.

Actual performance may vary depending on the task, workload and deployment environment.

Exa 3.0: AI-Powered Web Search

Exa 3.0 was highlighted as a search service designed for AI applications requiring access to fresh information from the internet.

The newsletter reported the introduction of a Lite pricing tier costing $4 per 1,000 searches.

Such services can help developers build applications that retrieve and process online information.

Bria Fibo 2.0: AI Image Generation

Bria Fibo 2.0 was presented as a downloadable AI image-generation model.

According to the newsletter, it ranks competitively among open image-generation models.

However, its free license does not permit commercial use.
Users intending to incorporate AI-generated images into commercial publications, advertising or other business activities should carefully review the applicable licensing terms.

The prices, capabilities, availability and licensing conditions of these tools may change over time.

AI Raises New Questions About Security, Transparency and Accountability

Taken together, these developments demonstrate the growing influence of artificial intelligence across multiple sectors.

AI-powered cyberattacks illustrate how advanced technology can be misused to threaten digital infrastructure.

The WhatsApp and Meta incidents raise questions about transparency, privacy protections and the responsibilities of companies operating AI-powered services.

Meanwhile, the rapid growth of AI-assisted shopping and the development of customized software show how artificial intelligence is creating new opportunities for businesses and consumers.

These changes are likely to influence how organizations protect information, interact with customers, manage transactions and develop digital products.

However, technological progress alone cannot guarantee public trust.

Strong cybersecurity safeguards, transparent AI systems, effective privacy protections and clear accountability mechanisms will be essential as artificial intelligence becomes more deeply integrated into everyday life.

Ultimately, the future of artificial intelligence will depend not only on how powerful these systems become, but also on how safely, transparently and responsibly they are developed and used.

Source Note: This report is based on information published in the Maze of Bot weekly AI newsletter, which referenced reporting and research associated with Reuters, Redact, The Verge, Upstarts Media and GitHub. The claims have not been independently verified by Nazrana Times.