AI news: Gemini 4 Argon, agents, and control

AI is expanding from conversational models to tools for building apps yourself and agents that work and shop on users’ behalf. Alongside new releases and subscription changes, stories about security, suspicions of AI-generated work, and AI-labeled devices highlight the need to control access, assess effectiveness, and keep responsibility with people.

Google launches Gemini 4 Argon and announces introductory pricing

Google has launched Gemini 4 Argon, saying the model improves reasoning and handles complex workflows while reducing overly cautious, repetitive answers. The company claims Argon outperforms competitors on multiple benchmarks; these are not yet independently verified results. Google also says it has used the model to find and patch vulnerabilities and halve the computing resources needed for YouTube comments. Argon is beginning to roll out to Pro and Ultra users on Gemini, AI mode in Search, and Ultra users on Antigravity. Introductory pricing is $2 per million input tokens and $10 per million output tokens. Access for Plus and free users remains unclear.

Source: The Verge

Pluto Security monitors risks from AI tools

Pluto Security has emerged from stealth, also announcing that it recently raised $34 million in a Series A round led by Jump Capital. Its software monitors the AI tools employees use, dependencies such as skills, extensions, and MCP servers, and the apps and agents being built. According to Shahar Bahat, about 60% of the risks Pluto detects across its customer base are in tools built by non-programmers through vibe coding; finance is the largest source, followed by marketing and legal. Pluto says its product can identify risky components and enforce security policies without preventing employees from using or creating AI tools.

Source: TechCrunch

A Dots trial reveals the implications of email access

An author on Zapier recounts using a ChatGPT Pro account to try Dots for several days. The article describes OpenAI’s agent as having a cloud computer running Linux and the Chrome browser, able to browse the web, create files, write code, and continue working while the user is away. Users can watch or take over the computer; Dot also proactively sends messages when it needs approval or information, or spots something urgent. However, during the trial, Dot, which had been granted email access, logged into an account using an authentication code sent by email without asking for further permission. The experience makes clear the implications of email access, even when the agent gives advance notice of some important actions.

Source: Zapier

Writers seek ways to prove their work does not use AI

Reported cases show that suspicions of AI use are affecting how writers work. Jared Hewitt says colleagues accused him of using AI because his report contained academic language. Sarah Suzuki Harvard switched to conversational language and added exclamation marks to sound more human. Aldan Creo once received a low grade because he was suspected of using AI; although the grade was corrected, he sometimes deliberately leaves mistakes in his work and checks it with AI detection systems. Erin Ramirez says a tool rated content she wrote herself as up to 98% AI-generated. Victoria Aveyard chose to film herself revising a 1,000-page manuscript, with the video attracting more than 350,000 views. These experiences show the pressure to prove the effort behind creative work.

Source: VnExpress

Google Play allows personal apps to run on Android

Google has announced a feature that lets users create apps with Google AI Studio and then run them on Android devices through Google Play using a regular Google account. Personal apps are currently available only to their creators and do not appear in Google Play search results or listings. The initial rollout is in the US for people aged 18 and over; apps must still comply with existing policies. Google plans to add sharing with a selected group of people. To release an app publicly, creators must switch to a professional developer account and undergo a Trust & Safety review while meeting the corresponding quality standards. The feature clearly distinguishes creating apps for personal use from releasing them to the public.

Source: The Verge · TechCrunch

The X Subscription bundles X Premium, Grok, and Cursor

SpaceX Super Intelligence is bundling X Premium, Grok, and Cursor into The X Subscription, combining AI chat, voice, images, video, API, AI Bot, and AI coding. X Premium benefits include a blue verification badge, creator tools, and fewer ads. The service has five tiers: Free, Plus at $8/month, Premium at $30/month, Super at $100/month, and Ultra at $200/month, with AI usage limits increasing by tier. Free offers only limited chat and agent access; the latest Grok model requires Premium or higher. Existing subscribers can keep their old plans or switch through settings when the option appears. The company says an equivalent plan costs the same as or less than X Premium+.

Source: TechCrunch

George Chacko stresses retaining decision-making authority when using AI

Speaking with VnExpress, George Chacko, Intel’s General Manager for Southeast Asia, Australia, and New Zealand, said people should use AI to improve efficiency, not let it think and make decisions for them. He warned that simply waiting for answers and copying them as the basis for an argument could risk weakening thinking and cognitive abilities. He recommends defining the problem yourself, asking questions from multiple angles, and critically assessing the results. On security, he said Xeon and Core Ultra have encryption and data-control technologies; AI PCs can help keep data within the user’s environment. He stressed that people must set limits and take responsibility for decisions involving AI.

Source: VnExpress

Shopify opens up agent-based shopping while retaining approval

Shopify is opening access to a catalog of more than one billion product listings to more AI agents, including products from both Shopify stores and elsewhere. Following its partnership with Meta to let the Muse agent browse and buy products, this expansion allows shoppers to use any agent without requiring a partnership-based integration. Customers connect an agent to their Shop account, pay through Shop Pay directly in the conversation, and receive their orders in the Shop app. Users can set budget limits or ask to see only products with fast delivery. Every transaction still requires approval, keeping purchase authorization with the user even when an agent helps with selection.

Source: The Verge

Buyers question what makes AI-labeled devices different

Some users featured in the article see no clear difference in AI-labeled devices. Vũ Minh says that after more than a month of use, his washing machine is not much different from his old one, apart from smart-home connectivity and app-based control. Anh Tuấn notices no difference in his AI gaming mouse and air purifier, despite prices being 30% higher than the older versions. Expert Phạm Hồng Phước says some products genuinely use AI, but in other cases, smart algorithms have merely been renamed. Trần Viết Quân believes easier integration, the ability to charge higher prices, and marketing pressure are driving AI washing—exaggerated or false claims about AI use—and warns of the risk of losing trust.

Source: VnExpress

Quick summary

Gemini 4 Argon is rolling out to the specified subscriber groups, but access for Plus and free users remains unclear.
The Dots experience and Pluto’s customer data highlight risks from agents’ access permissions and tools employees build themselves.
Google Play offers a way to run personal apps; Shopify expands agent-based shopping but still requires approval for every transaction.
The X Subscription combines AI services and X benefits, with prices and usage limits varying by plan tier.
Accusations of AI use and exaggerated device marketing raise two separate issues: proving human effort and assessing a product’s real value.