IT News Review by Control F5 Software: AI is decoding brain signals better and better.

AI decodes brain signals better and better

Research in the brain-computer interface area is entering a stage where AI can better interpret brain signals and transform them into text, intentions, or digital commands. The BBC Future material describes progress that combines AI models with data from EEG, fMRI, and more advanced neural interfaces, in an effort to reconstruct human language or intention more faithfully.

The immediate stakes are medical and accessibility-related. Such systems can help people who have lost the ability to speak or move to communicate more effectively, and the pace of progress suggests that multimodal models are becoming increasingly important in interpreting very noisy and difficult-to-map biological data.

For the software industry, the subject shows how quickly AI is moving from productivity to direct interface with hardware and sensitive data. With this evolution, questions related to privacy, consent, and the security of the infrastructure processing neural data are also increasing.

Perplexity bets on orchestrating multiple AI models

Perplexity has launched Computer, a cloud agent that promises to bring together multiple AI capabilities into a single workflow. According to TechCrunch, the product uses 19 different models, can create sub-agents for specific tasks, and is available on the Max subscription, priced at $200 per month.

The company claims that users need multiple models, not just one, because performance varies depending on the task. In the given examples, certain models are preferred for visual generation, others for software engineering or medical research, and Perplexity tries to automatically choose the optimal model for cost and accuracy.

From a product perspective, the signal is clear: differentiation no longer comes just from the base model, but from orchestration, routing, and user experience. For companies developing AI products, competition is increasingly shifting to the software infrastructure layer that decides when, where, and how each model is called.

Buddharoid brings AI into the ritual space

At Kyoto University, researchers presented Buddharoid, a humanoid robot trained on extensive volumes of Buddhist texts and built to provide spiritual guidance, answers to philosophical questions, and potentially support in religious rituals. The robot can speak, gesture, and adopt traditional prayer postures in temple spaces.

Euronews places the project in the context of an aging population and staff shortages in Japan. The system is led by Seiji Kumagai and combines advanced linguistic models with a commercial humanoid robot, aiming to go beyond previous religious robots that operated more on fixed scenarios.

For the software ecosystem, the case is relevant as it shows the expansion of conversational AI into highly specialized areas, where value comes not just from the response, but also from context, presence, and physical interaction. It is yet another example that the next stage of AI combines linguistic models, robotics, and experience design.

Robots and drones enter subway operations in China

A pilot system in Hefei shows how humanoid robots, robotic dogs, and drones can be integrated into the operations of an urban transport hub. At stations, robots guide passengers and respond to inquiries about transfers, while robotic dogs patrol for safety.

Meanwhile, autonomous robots inspect train components in maintenance technical rooms, using HD cameras and ultrasonic sensors to quickly identify cracks or loose parts. According to Euronews, the idea is for these systems to receive a common "brain" based on larger AI models in the future, to better respond to abnormal situations.

The practical implication is that AI and robotics are becoming increasingly relevant for critical infrastructure and predictive maintenance. For software teams, such projects require solid integration between computer vision, edge devices, orchestration, and alert systems that must operate in real-time and with low tolerance for error.

AI models frequently choose nuclear escalation in simulations

A preprint study cited by Euronews shows that in 95% of simulated war games, at least one model chose a form of nuclear escalation. The research placed ChatGPT, Claude, and Gemini Flash in the roles of leaders of nuclear superpowers, in a Cold War-type scenario.

The author, Kenneth Payne from King’s College London, claims that all three models treated tactical nuclear weapons as an additional step in escalation. According to the report, the models differentiated to some extent between tactical and strategic use, but the overall tendency remained an aggressive one.

For the industry, the important conclusion is that language models cannot be automatically treated as robust decision-making agents in sensitive contexts. In the area of software and AI safety, such results reinforce the need for simulation, red-teaming, and strict limitations on autonomy in systems that operate on critical or geopolitically impactful data.

Microsoft moves Copilot towards task-executing agents

Microsoft has introduced Copilot Tasks, a feature that allows the AI assistant to take on recurring, scheduled, or ad-hoc jobs based on natural language instructions. The Verge notes that the system can transform emails, attachments, and images into a slide deck or can periodically track real estate listings and even set viewings.

Copilot Tasks can also identify urgent emails, suggest responses, or manage personal activities, such as organizing a party. Microsoft positions the product as a response to the wave of agency capabilities launched by competitors like Claude, ChatGPT, Perplexity, or Google.

The direction is relevant for companies because AI is no longer presented just as a chat interface, but as an operational layer over applications, documents, and workflows. In practice, the real advantage will depend on how well authorization, observability, and execution control are resolved in enterprise environments.

Investors are no longer funding any AI SaaS

Interest in AI remains very high, but TechCrunch shows that investors are no longer attracted to any company that adds the "AI" label over a classic SaaS product. Instead, attention is shifting towards native AI infrastructure, vertical SaaS with proprietary data, systems of action products, and integrated platforms in critical workflows.

The message is that the market is starting to penalize superficiality. It is no longer enough to have a wrapper over an existing model or a good productivity demo if the product does not have unique data, deep integration, and clear operational value for the client.

For founders and product teams, this changes priorities. Differentiation must be built less from the interface and more from data, workflows, and infrastructure, which are exactly the areas where enterprise software can create real switching costs and sustainable advantages.

Honor pushes further the segment of thin foldable phones

Honor launched the Magic V6 ahead of MWC, emphasizing a very thin format and a 6,600 mAh battery. TechCrunch notes that the device is 4 mm thick when unfolded and 8.75 mm when folded, plus a newly designed hinge for better durability.

Although the thickness differences compared to the previous generation are small, the company continues to use industrial design as a central positioning element. In the premium segment, such launches show that hardware innovation is increasingly shifting towards optimizing the trade-off between battery life, durability, and form factor.

For the digital ecosystem, the relevance comes from the fact that foldables are becoming a more mature platform for adaptive software experiences. As hardware stabilizes, the real difference will come from applications and interfaces that know how to leverage the flexible screen and multitasking.

Xiaomi bets on photography, accessories, and ecosystem

Xiaomi announced a series of new products at MWC, from the flagship smartphone 17 Ultra to a tracker compatible with Apple Find My and Google Find Hub, plus a very thin magnetic power bank. The phone is co-branded with Leica and emphasizes the main 50 MP camera and the 200 MP telephoto lens with variable optical zoom.

The company also presented dedicated photography accessories, including kits that add physical buttons, zoom control, and additional battery. Meanwhile, Xiaomi is expanding availability in the EU and the UK, with a starting price of 1,499 euros for the 17 Ultra and 1,999 euros for the Leica edition.

The message is that the premium smartphone is no longer sold as a standalone product, but as a node in an ecosystem of accessories, services, and software integration. For tech players, the model shows how important it becomes to combine hardware with extensions that amplify concrete use scenarios.

ChatGPT approaches the threshold of one billion weekly users

OpenAI announced that ChatGPT has reached 900 million active weekly users and 50 million paying subscribers. The company says that the beginning of the year has accelerated the growth rate of subscribers, and the product is improving as the scale increases.

The data confirms that generative AI has surpassed the curiosity stage and has entered the category of mass products with recurring use. At this point, the discussion is no longer just about adoption, but about retention, monetization, and how the infrastructure supports such large volumes.

For business and software, the signal is that conversational interfaces are becoming a standard channel, not an experimental one. Any company building digital products must already consider integration with such flows or indirect competition with them.

Suno crosses important monetization thresholds

The music generator Suno has reached 2 million paying subscribers and 300 million dollars in annual recurring revenue, according to information provided by co-founder and CEO Mikey Shulman. TechCrunch notes that just three months ago, the company announced a funding round of 250 million dollars, at a valuation of 2.45 billion.

The jump from 200 to 300 million dollars ARR in such a short time suggests real demand for vertical AI products when the outcome is clear and immediate. In Suno's case, the product does not just sell technology, but a directly monetizable creative experience.

For the market, this is a signal that certain consumer AI applications can build substantial businesses without necessarily starting from the enterprise space. At the same time, the pressure on infrastructure, licensing, and copyright will increase as such platforms delve deeper into the mainstream.

OpenAI attracts 110 billion dollars in a historic round

OpenAI announced a private funding round of 110 billion dollars, one of the largest in history, at a pre-money valuation of 730 billion dollars. According to TechCrunch, the announced contributions come from Amazon, Nvidia, and SoftBank, and the round remains open for other investors.

The size of the transaction shows that AI has entered a stage where competition is supported not only by research and product but also by extreme financial power. The costs for compute, models, partnerships, and distribution are pushing the ecosystem towards a sharper concentration.

For software companies, this type of round indicates that the global AI infrastructure will be increasingly controlled by a small number of players with access to capital, chips, and distribution. Differentiation for the rest of the market will come mainly from products, proprietary data, and specialization.

The memory crisis could hit the smartphone market hard

The increased demand for RAM for PCs and AI-dedicated data centers is creating major pressure on the supply chain, and IDC estimates that global smartphone shipments could decrease by 12.9% in 2026. TechCrunch says this would be the largest annual decline in the last decade.

IDC anticipates a drop from 1.26 billion units in 2025 to 1.12 billion in 2026, along with a 14% increase in the average selling price. The low-end segment is the most exposed, and phones under 100 dollars could become structurally unviable.

For the industry, the subject shows how the AI boom can destabilize seemingly separate hardware categories. When memory becomes a strategic resource, the effects propagate from data centers to volume products, forcing vendors to rethink their portfolios, launch schedules, and specification trade-offs.

Read AI launches a "digital twin" via email

Read AI has introduced an email-based digital twin, designed to respond on behalf of the user to inquiries and assist with scheduling and coordination. The TechCrunch article positions the product as a new step in the direction of AI agents that operate through channels already used daily, not just through separate applications.

The idea of a digital twin suggests an assistant that knows the context of conversations, the agenda, and the user's preferences well enough to mediate certain interactions. Practically, email becomes an interface for automation, not just a communication medium.

For companies, the relevance is high because email remains a critical layer of operations. Any agent working here must solve permission, audit, and traceability issues very well; otherwise, the productivity benefit may come with high operational risk.

Instagram expands parent alerts for risk searches

Instagram has introduced alerts for parents when teenagers search for content related to suicide or self-harm. The TechCrunch article indicates a new stage in Meta's approach to minor safety, by combining monitoring of search behaviors with notifications to families.

The move shows that social platforms are trying to shift protection from being exclusively reactive, based on reports or post-publication moderation, to a more preventive model. This is an important change as it emphasizes the context of use, not just the published content.

For digital products, the theme highlights how difficult it is to balance safety, privacy, and autonomy. Implementing such features requires content classification, behavioral signals, and intervention rules that must be calibrated very carefully.

Gemini automates multi-step tasks on Android

Google is expanding Gemini on Android with automation capabilities for certain tasks in multiple steps. According to TechCrunch, the direction brings the assistant closer to an agent capable of executing, not just responding, within the mobile ecosystem.

This type of functionality transforms the phone into a space where AI can orchestrate applications and successive actions. It is an important move because the mobile has remained more limited than the desktop in terms of complex automation based on natural language.

For developers, the implication is that integration with AI assistants at the operating system level is becoming increasingly relevant. Applications that offer clear endpoints, well-defined permissions, and predictable flows will be easier to connect in such agentic scenarios.

OpenAI treats advertising as an iterative experiment

OpenAI's COO, Brad Lightcap, stated that the introduction of ads will be an iterative process and that announcements can add value to the experience if done correctly. TechCrunch notes that the executive practically asked for time to see how the product evolves in the coming months.

The message confirms that monetizing ChatGPT will not rely solely on subscriptions but may also include advertising over time, although in a still unclear form. In the context where user trust is critical for a conversational AI product, this transition will be closely monitored by the market.

For AI-based software products, the subject is important as it raises an old question in a new context: how to monetize without compromising perceived utility and neutrality. For chatbots and assistants, this balance is even more sensitive than in social media or search.

Alexa+ receives new personality options

Amazon has introduced three new personality styles for Alexa+, named Brief, Chill, and Sweet, which modify the tone of the assistant's responses. TechCrunch presents the change as an extension of the idea that the AI voice is no longer just utilitarian but also a product personalization element.

The feature does not necessarily change what Alexa+ can do, but it changes how the relationship with the user is shaped through tone, rhythm, and style. This shows how much competition among assistants is shifting towards experience and individual preferences.

For product teams, the direction is relevant because personality becomes a configurable parameter of the interface. As models standardize, how they respond, not just what they respond, can become a real commercial differentiator.

Instagram also expands the TV app on Google TV

Instagram is bringing its TV app to Google TV devices in the US, two months after its debut on Amazon Fire TV. The app was initially launched to extend the consumption of Reels beyond mobile, and now it also allows browsing through posts in the feed directly on the TV.

The move indicates a clear interest in moving short vertical content into a space previously dominated by YouTube and traditional video services. Social networks are thus trying to reclaim consumption time in the living room, not just on the phone.

For the digital industry, this expansion shows how the boundaries between social, video platforms, and TV OS are becoming increasingly permeable. With distribution on televisions, the format, recommendations, and monetization of short content enter a new stage.

Nvidia is preparing a new chip for AI acceleration

According to a report picked up by Yahoo Finance from the Wall Street Journal, Nvidia is working on a new processor designed to help clients like OpenAI build faster and more efficient AI systems. The information comes at a time when competition for AI infrastructure is becoming increasingly intense.

Even though public technical details are limited, the direction is consistent with Nvidia's strategy to control not only training acceleration but also as much of the inference and optimization stack as possible. The demand for efficiency, not just raw performance, is becoming increasingly important as volumes grow.

For the software market, each hardware leap changes the calculations related to cost per token, latency, and application architecture. Therefore, chip-level innovation continues to have a direct effect on AI products delivered to businesses and consumers.

Meta faces a difficult road in the EU antitrust dispute

An advisor to the EU Court of Justice has issued an opinion supporting broad requests for digital evidence in antitrust investigations, suggesting that Meta's appeal has little chance of changing the current standard. Courthouse News describes the situation as a serious obstacle for the company in its dispute with European authorities.

The stakes are not just for Meta's case but also how extensive the investigative tools can be against large digital platforms. If the standard remains permissive for authorities, tech companies will operate in an increasingly strict compliance and documentation framework.

For the software and data area, the case reconfirms that information governance is becoming part of business infrastructure. As products become more integrated and data volumes grow, so do legal risks and audit requirements.

Schneider Electric grows amid demand from data centers

Schneider Electric has exceeded profit estimates, supported by demand from the data center sector. Yahoo Finance notes that the company anticipates organic revenue growth between 7% and 10% for 2026 and an improvement in adjusted EBITA margin by 50 to 80 basis points.

The context is relevant because the physical infrastructure of the AI boom—power, cooling, electrical distribution, and industrial automation—becomes as strategic as chips and models. The demand for data centers translates not only into revenues for hyperscalers but also for critical equipment suppliers.

For companies, the story shows that AI is not just a software wave. It is also an industrial, energy, and operational wave, and those building AI applications increasingly depend on the robustness of a physical infrastructure that must scale rapidly.

Experity automates administration in urgent care

Experity claims that its AI agent for urgent care has saved 7,000 hours of administrative work by automating follow-ups and reducing operational burden. The product is an example of AI applied to a very concrete problem, not just a generic productivity flow.

In healthcare, much of AI's value does not come from spectacular models but from eliminating repetitive tasks that consume clinical and administrative time. Right here, a well-integrated agent can produce ROI faster than in areas where accuracy requirements are harder to control.

For the software industry, the case shows the potential of vertical applications with deep integration into workflows. In regulated sectors, the gain comes from pragmatic automation, traceability, and reducing friction, not from abstract promises about "AI for all."

Lenovo tests a dedicated AI assistant for the workplace

At MWC 2026, Lenovo presented the AI Work Companion concept, described by Notebookcheck as an evolution of the former Smart Clock Essential, but focused on productivity and hub functionality at the office. The device moves the idea of an AI companion from the bedroom to the desktop.

The concept suggests that manufacturers are looking for new hardware formats to integrate AI into daily work without limiting it to a laptop or phone. Such a device could become a dedicated interface for notifications, shortcuts, contextual control, and integration with other systems in the workspace.

For the market, it is yet another signal that AI is seeking its own hardware form, one that combines constant presence with quick access to digital services. If these concepts become real products, productivity software will need to adapt to new interaction points.

Apple may replace Core ML with a new Core AI framework

9to5Mac reports that Apple is preparing a modernized framework called Core AI at WWDC, which would replace Core ML in iOS 27. The goal would be to help developers better utilize modern AI capabilities in their applications.

If this change is confirmed, Apple conveys that the current stack for machine learning needs to be updated for an era dominated by generative models, agents, and more flexible on-device inference. It is no longer just about classification or predictions, but about a new set of primitives for native AI applications.

For developers, the implication is significant. A platform-level framework change can redesign how iOS applications integrate local models, accelerate AI tasks, and connect to hardware capabilities in the Apple ecosystem.

Google Translate uses Gemini for contextual alternatives

Google Translate is expanding the use of Gemini to provide alternative formulations based on context, not just a singular translation. The Verge notes that the app now offers multiple expression variants on iOS and Android, plus new buttons like Understand and Ask for additional context.

The direction reflects an important change in linguistic products. Translation is no longer treated as a mechanical mapping between two languages but as contextual interpretation, where tone, register, and idiomatic meanings matter.

For global software products, this matters directly. Tools that succeed in modeling nuance, not just literal meaning, can significantly improve the experience in support, e-commerce, content, and international collaboration.

NATO approves iPhone and iPad for classified information at "restricted" level

iPhone and iPad have been approved to manage NATO classified information at the restricted level. According to Apple, commercial devices with iOS 26 and iPadOS 26 can do this without special software or additional settings.

The announcement is notable because it validates the security of consumer devices in a very demanding institutional framework. In a period where mobility and access to classified data intersect more frequently, such certifications change the discussion about BYOD and about the endpoints allowed in sensitive organizations.

For companies, the subject shows how important the platform's security layer is, not just the applications running on it. When a commercial device receives such approval, the impact is felt in procurement, access policies, and the design of enterprise mobile applications.

Google democratizes advanced image generation features

Google is launching Nano Banana 2, also known as Gemini 3.1 Flash Image, to free users from Gemini and other company AI platforms. The update further brings capabilities that were previously reserved for Nano Banana Pro, including generation with readable text and localized translation in images.

The company positions the model as a combination of speed, low cost, and better visual capability. Essentially, Google is trying to bring professional-quality tools to a wider access layer, including in Search AI Mode.

For the market, the trend is important as it reduces the barrier between premium tools and mass ones. As advanced features become free or nearly free, differentiation will depend less on access and more on integration, workflow, and specialization.

The YouTube algorithm frequently pushes "AI slop" to children

After watching popular children's channels like CoComelon, Bluey, or Ms. Rachel, journalists cited by The Verge observed that over 40% of recommended Shorts appeared to contain AI-generated visual elements. The problem is amplified by the fact that YouTube does not require explicit labeling of these animations for children.

This shifts the entire burden of filtering onto parents and reveals the current limits of automated moderation on high-volume platforms. It is not just a problem of low-quality content but also one of trust in recommendation systems when the audience is very vulnerable.

For companies building feeds and recommendation engines, the subject is highly relevant. The quality of the classification model and labeling policies becomes as important as engagement, especially in areas with high sensitivity.

Google brings the Intrinsic project closer to the core

Google is integrating Intrinsic, the AI robotics project within Alphabet, directly into the company. The move is a sign that the giant wants to bring "physical AI" projects closer to its core business and reduce the distance between experimentation and commercialization.

Intrinsic had been viewed by many as a kind of "Android of robotics," a software layer meant to simplify the programming and operation of industrial robots. By bringing it closer to Google, the company seems to want more control and a more connected execution to the rest of its AI ecosystem.

For the market, the signal is that robotics is back in the spotlight, but this time with generative models and cloud infrastructure as central elements. For software teams, this means opportunities in orchestration, simulation, control, and integration between AI and physical systems.

Reddit gains ground with AI search

Reddit has become increasingly present in AI-generated search results, amid its partnership with Google and its growing visibility in search starting in 2024. A major factor is that the platform provides conversational, up-to-date, and often very specific content, useful for summarization and contextual response.

The platform also has agreements with other players, including OpenAI, which places it in a strategic position in an internet increasingly mediated by AI systems. What was previously seen as community noise is starting to be treated as valuable material for automatically generated responses.

For business and publishing, the change is significant. Visibility on the web no longer depends solely on classic SEO but also on the content's ability to be extracted, cited, and reinterpreted by AI systems that favor concrete human experiences and responses.

More and more American teenagers use AI for emotional support

16% of teenagers in the US use AI for informal conversations, and 12% use it for emotional support or advice. At the same time, the most common uses remain information searching and school help.

The data suggests that chatbots are starting to occupy roles that until now belonged to friends, family, or counselors. Even though usage is not majority, the percentage is high enough to raise serious questions about dependency, validation, and the quality of the advice offered.

For companies building conversational AI products, the subject underscores the responsibility of design. When users seek emotional support, safety, limits, and escalation mechanisms are no longer optional functions.

Scenarios about an AI-driven economic collapse enter the mainstream

A Euronews piece presents a thought experiment imagining a "global intelligence crisis" by 2028, generated by the accelerated replacement of high-skilled labor with AI agents. The text starts from a fictional memo written from the perspective of June 2028, signed by James Van Geelen and Alap Shah.

The scenario describes a vicious cycle where layoffs reduce consumption, affecting the economy and amplifying social tension. It is not an official forecast but a strategic imagination exercise meant to raise questions about the pace at which AI can replace certain categories of work.

For the business environment, the utility of such material does not lie in the accuracy of the timeline but in the questions it raises about organizational adaptation, reskilling, and transition design. In the absence of these, the productivity brought by AI may come with social costs that are hard to absorb.

Barcelona tests robots for assistance and companionship at home

Barcelona is running a public project where 1.35-meter robots are sent to the homes of individuals in the early stages of cognitive decline. Resident Irene Veglison, aged 67, received the robot in November, and the project aims to improve tele-assistance.

Local authorities say that in the future, robots could detect risks and alert professionals, for example, if a person has fallen and cannot respond. The context is familiar in Europe, where increasing life expectancy and declining birth rates are putting more pressure on care systems.

For the software and AI industry, the case shows how quickly domestic assistance is becoming a practical category, not an experimental one. The real value lies not just in conversation but in event detection, integration with human services, and reliability in real environments.

Google prepares changes in Search under DMA pressure

According to a Reuters report, Google is ready to test changes to search results to provide more visibility to competitors, in an attempt to avoid a fine from the EU. The measure specifically targets searches in areas such as hotels, flights, and restaurants.

The changes would display both Google results and results from vertical search services, and the highest-ranked such services could appear by default. It is an important concession in a broader conflict related to the Digital Markets Act and the alleged favoritism of its own services.

For the digital business, the impact could be significant. Any change in how Google presents results affects visibility, traffic distribution, and SEO strategies, especially for platforms that rely on search mediation.

The silent risk of AI is starting to concern companies more

A CNBC piece, summarized in other accessible sources, emphasizes the idea of "silent failure at scale," meaning situations where AI systems do not fail spectacularly but discreetly, and errors accumulate until they become a serious operational problem. Alfredo Hickman from Obsidian Security says that organizations are pursuing a moving target, as even developers do not clearly know how technology will evolve in the coming years.

As AI is connected to transaction approvals, code writing, customer interactions, or data transfers, the gap between expected behavior and actual behavior in production grows. Noe Ramos from Agiloft warns that these systems do not always fail loudly but sometimes make gradual mistakes over long periods until the impact becomes visible.

For software companies, this is likely one of the most useful warnings of the moment. The central issue is no longer just whether a model works in a demo, but whether it can be monitored, audited, and shut down in a real, complex system connected to business processes.

Synthesis made with the help of a monitoring flow provided by Control F5 Software.