IT News Review by Control F5 Software: Spain launches a new tool for measuring hate in social media

Spain launches a new tool for measuring hate on social media

The Spanish government is preparing HODIO, a system designed to measure hate speech on digital platforms and assess the role of algorithms in amplifying polarization. The announcement was made by Prime Minister Pedro Sánchez at the international summit against hate and digital harassment in Madrid.

Euronews notes that the system will analyze large volumes of public activity on social networks to measure the intensity, reach, and evolution over time of hate speech. The government says the goal is twofold: to build more effective public policies and to increase pressure on technology platforms to take more responsibility.

For the software area, the initiative shows how social media observability is becoming a public infrastructure issue, not just a commercial analytics one. Tools that combine large-scale analysis, temporal measurement, and evaluation of algorithmic effects could become increasingly relevant in European digital policies.

Nearly 40 startups have become unicorns in just a few months

Nearly 40 venture capital-backed startups have already reached unicorn status by 2026, amid a new investment frenzy largely fueled by AI. Although many of the companies are in the AI space, startups are also emerging in health, cloud infrastructure, energy, and crypto.

Examples include companies like Positron, an AI semiconductor startup, Apptronik, focused on humanoid robots, Oxide, which builds cloud infrastructure for private data centers, and Profound, an SEO product designed for brand visibility in AI engine responses. The list also includes companies in telemedicine, home batteries, digital psychiatry, and accounting automation software.

For the software market, the signal is clear: investments are not just going towards models and chatbots, but towards the entire infrastructure layer, vertical applications, and enterprise products that can leverage the AI wave. The fact that TechCrunch is already seeing nearly 40 new unicorns this early in the year suggests that investors are quickly looking for not just promises, but also clear positioning in the value chain of the new digital economy.

The EU wants to ban AI tools for generating non-consensual sexualized images after the Grok scandal

In Europe, an initiative is taking shape to ban AI tools that can generate non-consensual sexualized images of real people. The proposal comes after controversies surrounding Grok, the chatbot integrated into X, which has been associated with generating such images and has already attracted European investigations regarding illegal content and platform safety.

According to reports on the proposal, the text on the table of EU diplomats would ban the promotion of any AI tool capable of creating images, videos, or audio with intimate parts or explicit sexual acts involving identifiable individuals, without consent. The material also states that the amendment would be made through adjustments to the AI Act, and enforcement could come as early as this summer.

For companies developing generative products, the case shows how quickly new restrictions can arise when a technical capability comes into direct conflict with user safety and fundamental rights. Practically, this means more pressure on guardrails, output moderation, and clearly defining commercial boundaries for multimodal models.

ByteDance halts global launch of Seedance 2.0 after copyright disputes

ByteDance has suspended the global launch of its AI video model Seedance 2.0, following a series of copyright disputes. SCMP and other subsequent reports say the move comes after complaints and cease-and-desist letters from major Hollywood studios.

Reuters notes that ByteDance had promised back in February measures to prevent unauthorized use of intellectual property in Seedance 2.0, following pressure from Disney. In the same context, viral clips generated by the model circulated online, including scenes with Tom Cruise and Brad Pitt, and studios accused the company of using characters and protected materials without permission for training and operation.

For the AI industry, the ByteDance case is significant because it shows how quickly a spectacular technical product can run into legal blockages if training sources and protection mechanisms are not sufficiently clear. In the generative video space, copyright compliance is already becoming a launch and distribution issue, not just a secondary risk.

MacBook Neo becomes the most repairable Apple laptop in over 10 years

The new MacBook Neo is the most repairable laptop launched by Apple since 2012, according to an iFixit analysis. The laptop, announced at the beginning of the month and positioned as an entry-level model, starts at $499 for students.

An analysis highlighted several important changes compared to previous generations: the battery and keyboard are attached with screws, not adhesive or rivets, and components like the camera and fingerprint sensor can be replaced more easily. However, the final score was only 6 out of 10, among the reasons cited being the DRAM memory soldered directly onto the motherboard, making upgrades difficult.

From a software and hardware perspective, the news is relevant because it shifts the discussion about the Apple ecosystem beyond performance and price. Repairability is becoming increasingly important in purchasing decisions, lifespan, and total cost of ownership, especially in education and in organizations that manage large fleets of devices.

Travis Kalanick returns with Atoms, a company focused on specialized robots

Travis Kalanick has launched a new company called Atoms, focused on robotics for industries such as food, mining, and transportation. TechCrunch reports that the current company CloudKitchens will be integrated into the new structure.

According to the public description, Atoms will build a "wheelbase for robots," and Kalanick has stated that he is targeting specialized robots, not humanoids. In the same interview, he also indicated that he is approaching the acquisition of Pronto, the autonomous vehicle startup for industrial and mining sites created by Anthony Levandowski, adding that he is already the largest investor in the company.

For the industrial software market, the message is clear: the new wave of robotics is not limited to humanoid demonstrations, but is moving towards integrated systems for specific use cases. Where hardware is coupled with autonomy, orchestration, and operational software, economic value can be more easily demonstrated and monetized.

Nvidia aggressively pushes CPUs in a new phase of AI

Nvidia is shifting a significant part of its discourse and offering towards inference and the role of CPUs in the new phase of the AI market. At GTC 2026, the company stated that the revenue opportunity for its AI chips could reach at least $1 trillion by 2027, double the previous estimate that capped at $500 billion by 2026.

In the same context, Jensen Huang presented a new CPU and an AI system based on Groq technology, emphasizing that Nvidia wants to compete more aggressively in inference, where its GPUs are increasingly facing CPUs and custom processors built by players like Google. Reuters explicitly notes that the strategy aims to run AI in real-time, not just train models.

For the industry, this is an important sign that the AI stack is becoming more diverse and optimized for workload. As agents and live applications grow, it will not only be about who has the most powerful accelerator, but also about who can effectively combine CPU, GPU, network, and energy consumption into a coherent architecture.

Meta expands its AI chip family and raises pressure on Nvidia and AMD

Meta has presented a roadmap with four new AI chips developed in-house, as part of the Meta Training and Inference Accelerator program. Reuters reports that the first of these, MTIA 300, is already being used for the company's ranking and recommendation systems.

The other chips will be launched gradually in 2026 and 2027, and the MTIA 450 and 500 models are specifically designed for inference, meaning the execution of responses in AI applications. Reuters also notes that starting with MTIA 400, Meta is designing the system around the chip, including rack-type components and liquid cooling, in a clear effort to reduce costs and energy consumption compared to external solutions.

For the software and infrastructure ecosystem, the stakes are high. As hyperscalers develop more of their own chips, the AI stack becomes more vertically integrated, and the competitive advantage shifts from simply acquiring hardware to controlling the entire platform delivery, from model to datacenter.

OpenAI wants to bring Sora directly into ChatGPT

OpenAI intends to soon launch the video tool Sora in ChatGPT. Until now, Sora has operated as a separate application, launched in September 2025.

The standalone application allows for the creation and distribution of AI videos, including clips derived from protected content, published in a social media-like feed. According to reports, OpenAI will not shut down the standalone application but will maintain it in parallel with the integration into ChatGPT.

For product and distribution, the move makes strategic sense. Integrating Sora into ChatGPT would bring video generation into an already massively adopted interface and reduce friction between conversation and visual creation, transforming ChatGPT into a more complete hub for multimodal content.

Substack adds a recording studio directly into the platform

Substack has launched Recording Studio, an integrated tool that allows creators to record and publish videos directly from the platform. The feature is available on desktop and supports both solo recordings and conversations with up to two guests.

The platform states that the new studio includes customizable watermarks, screen sharing, and automatic generation of clips and thumbnails after the recording is completed. In the company's communication, Substack argued that until now creators needed a separate set of tools for filming, distributing clips, and designing promotional materials.

The news is relevant for the creator economy and publishing software as it shows how platforms are trying to internalize as much of the workflow as possible. The more steps are solved in a single product, the greater the retention of creators and the value of the platform as media infrastructure.

Bumble introduces Bee, an AI assistant

Bumble has announced Bee, a new AI assistant described as a personal matchmaker that infers from private conversations the values, communication style, lifestyle, and dating intentions of users. Currently, the product is being tested internally and is set to enter beta.

The company says that Bee will initially power a new experience called Dates, in which the AI recommends matches based on compatibility of values and goals. TechCrunch notes that Bumble sees Bee as a way to collect more context about users, at a time when the dating app market is facing fatigue from the Gen Z audience.

For consumer products, the launch shows how AI is being pushed beyond simple suggestions and into the realm of relational mediation. This can increase personalization but also raises questions about how much personal data a platform needs to accumulate to deliver better matching.

Grammarly is sued for using authors' names in "Expert Review"

Following criticism regarding the Expert Review feature, Grammarly is also facing a lawsuit. Journalist Julia Angwin has filed a class action against the parent company Superhuman, claiming that the product used her name and those of other authors without consent to simulate AI-generated editorial feedback.

The controversial feature promised suggestions that appeared to come "from the perspective" of well-known individuals, such as Stephen King, Carl Sagan, or Kara Swisher. According to the article, the lawsuit invokes violations of privacy and image rights, and Angwin states that it is about an impostor version of her professional expertise.

For the SaaS industry, the case is important because it shifts the discussion from UX and naming to legal issues. As AI products increasingly approach the simulation of recognizable identities and styles, the boundary between inspiration, representation, and commercial use without consent becomes more sensitive.

Alexa+ receives a "for adults" personality, with sassier replies

Amazon is expanding the range of personality styles for Alexa+ with an option called Sassy, intended exclusively for adults. Activating this personality requires additional verifications in the Alexa app, and the feature is not available when Amazon Kids is activated.

The new option joins the previously introduced Brief, Chill, and Sweet styles. According to the description provided by Amazon, Sassy uses explicit language and an ironic tone, with responses "help first, judge always," but without entering the NSFW content area.

For conversational products, this type of personalization shows that differentiation is no longer just about what an assistant knows, but also about how it speaks. The tone thus becomes an explicit component of the product, with direct implications for segmentation, safety, and commercial positioning.

WordPress launches a private space that runs directly in the browser

WordPress can now run entirely in the browser through a new service called my.WordPress.net. Users can create a site and start publishing without a traditional hosting account, without domain setup, and without the usual installation process.

The solution uses the same technology that powered WordPress demos until now, but is transformed into a permanent personal platform. However, there is an important limitation: sites created this way are private by default and not accessible on the public internet. WordPress describes the product as a personal environment where ideas can exist before they are ready for publication or even before they are ever published.

For the software ecosystem, this is a good example of shifting complexity from infrastructure to user experience. As publishing and prototyping become easier to start, directly in the browser, the technical barrier for testing, drafting, and editorial experimentation decreases.

TikTok allows Apple Music subscribers to listen to full songs without leaving the app

TikTok has entered into a partnership with Apple Music that allows subscribers of the service to play a full song without leaving the TikTok app. The new option appears through the Play Full Song button in the feed or on the audio detail page.

The feature is built on MusicKit, meaning that full playback occurs through Apple Music, and streams are paid within that service. The publication also emphasizes that TikTok does not offer the same experience for other streaming platforms, including Spotify.

For the digital product space, the partnership is relevant as it further bridges content discovery with monetization. TikTok strengthens its role in the music ecosystem, while Apple Music gains a direct entry point into one of the strongest discovery engines.

The European Parliament calls for new rules to protect works used for AI training

The European Parliament has adopted a series of recommendations calling for a permanent solution to protect copyright in the context of training AI systems. Euronews reports that the report is seen by creative organizations as a strong political signal of support for artists and rights holders.

Among the proposals is a European registry, hosted at EUIPO, to list all protected works used to train models and the artists who have opted not to allow this use. The report also calls for companies to disclose which sites have been scraped for training data and warns that lack of transparency could equate to copyright infringement.

For companies in AI and software, the direction is clear: traceability of training data and remuneration for creators are becoming structural issues, not legal exceptions. As transparency requirements increase, the provenance of data becomes part of product strategy and compliance.

Google Maps receives Ask Maps and more immersive navigation with Gemini

Google is bringing Gemini to Google Maps through a new feature called Ask Maps, designed for complex and very specific questions about places, routes, and plans. The new system can respond conversationally and personalized, using context from the user's previous interactions with Maps.

Examples given include detailed requests, such as finding a vegetarian restaurant in a specific area, with a specific atmosphere and a table available at a certain time. Google states that the responses are based solely on data from Maps, such as reviews, photos, saved or favorite places, and that personalized results are not influenced by paid placements. The feature is starting to roll out in the US and India on Android and iOS.

For the software area, Ask Maps shows how AI is moving from generic search to transactional and contextual interfaces. When a product can understand intent, filter its own sources, and transform the response into action, the difference between search, recommendation, and execution begins to disappear.

Canva tries to make AI images easier to edit, not just generate

Canva has launched in public beta Magic Layers, a feature that transforms flat images and AI-generated visuals into designs with editable layers. The feature is available in the US, UK, Canada, and Australia.

Magic Layers can separately select objects, text boxes, and other graphic elements, while maintaining the original layout. Canva states that the idea is to eliminate the need to re-prompt an image just to adjust a small area. The feature currently supports PNG and JPEG files with a single page, not just AI-generated content.

For digital creators and creative software, the novelty is important as it shifts the focus from generation to editability. Practically, tools that offer fine control after generation are more likely to become useful in real workflows than those that deliver only spectacular but rigid results.

AI and software agencies may put significant pressure on hiring graduates

In a cited intervention by CNBC and picked up by other publications, ServiceNow CEO Bill McDermott stated that unemployment among new graduates could rise to 30% in the coming years due to automation produced by AI agencies. He linked this development to the increasing disappearance of routine tasks traditionally assigned to entry-level roles.

Fortune notes that McDermott said that about 9% of recent college graduates are already facing unemployment and suggested that digital agencies could take over much of the undifferentiated work. The article also mentions data from the New York Fed, according to which the unemployment rate for recent graduates is already above the overall average.

For companies and the job market, this is an important signal about the secondary effects of AI. It is not just about operational efficiency, but also about what happens to the roles through which young people enter organizations, gain experience, and build professional experience.

AI data centers are increasingly at the center of disputes over electricity bills

Amid the rapid growth of AI infrastructure, pressure on electrical grids and consumer bills has become a major political and economic issue in the US. Reuters and E&E News show that the Trump administration and the utility industry are trying to convey that new data centers should not push costs onto households.

In this context, the so-called Ratepayer Protection Pledge has emerged, through which large tech companies promise to bring or purchase electricity sources for their own data centers, pay for upgrades to delivery infrastructure, and enter into special rate agreements. Reuters notes, however, that there is also skepticism from some observers, who say that paying for new capacities does not automatically mean they will be brought online quickly enough to relieve pressure on the grid.

For the software and AI industry, the dispute is relevant because it shows that scaling infrastructure is no longer just a technical issue. Computing power is becoming increasingly tied to regulation, public acceptance, energy costs, and the relationship between hyperscalers, utilities, and authorities.

China approves the first commercially authorized brain implant

China has approved the first brain implant for commercial use intended for people with paralysis, and Euronews describes it as the first commercial authorization of its kind worldwide. The device is developed by Neuracle Medical Technology.

The system uses a brain-computer interface to detect signals associated with the intention to move the hand and translates them into commands for a robotic hand. This opens and closes with pneumatic assistance, allowing the user to grasp objects. The technology is intended for individuals with severe spinal cord injuries in the neck area.

For the software and AI area, the news is important because it shows the maturation of products that combine signal modeling, real-time interpretation, and assistive hardware. As brain-computer interfaces move out of the lab and into commercial use, the pressure on reliability, integration, and safety will increase significantly.

Google uses millions of old articles to predict flash floods

Google says it has found a new way to improve flash flood forecasting by using Gemini to read millions of press articles. These phenomena kill over 5,000 people annually and are very difficult to anticipate precisely because of their local and short-lived nature.

The company's researchers analyzed 5 million articles from around the world and isolated 2.6 million distinct floods, transforming the information into a geo-tagged time series called Groundsource. Google states that this is the first time it has used a large language model in this way to compensate for the lack of classical measurement data.

For the AI industry, this is an interesting example of using language models beyond conversation and generation. Practically, LLMs are becoming a tool for extracting and structuring signals from chaotic textual sources, which can expand access to useful data in areas where sensors and historical bases are insufficient.

AI applications have long-term retention issues

Simply integrating AI into an application does not guarantee good retention of subscribers. TechCrunch notes that AI applications have, on average, a churn rate 30% faster for annual subscriptions than non-AI applications.

The study is based on data from developers using RevenueCat to manage over 1 billion in-app transactions and over $11 billion in annual revenue. The sheer size of this sample makes the conclusions relevant for the subscription app ecosystem on iOS, Android, and the web.

For software companies, the conclusion is useful: AI can help with acquisition and initial curiosity, but it does not automatically solve the problem of perceived value over time. Retention continues to depend on repeated utility, price, experience, and good integration into user habits, not just the "AI-powered" label.

An AI model can detect the risk of domestic abuse years before victims seek help

Researchers in the US have developed an AI system designed to help doctors identify patients at risk of intimate partner violence even years before they seek support. Euronews notes that current screening in hospitals often relies on direct questions, and many victims do not disclose abuse due to fear, stigma, or safety reasons.

The model was built on several years of medical data from nearly 850 women who have gone through such situations and over 5,200 similar patients from the control group. Three systems were tested, and the combined model, which analyzed both structured data and medical notes and radiology reports, correctly identified risk in 88% of cases and could signal abuse more than three years before many patients entered intervention programs.

For digital health and applied AI, this is a powerful example of using existing data for early intervention. At the same time, the subject requires a lot of caution: any such system must be integrated with very clear rules of confidentiality, clinical validation, and responsible human intervention.

"AI brain fry" describes the mental fatigue caused by intense work with chatbots

A study cited by Euronews suggests that intensive use of AI chatbots at work can produce a form of cognitive exhaustion called "AI brain fry." The research, conducted on over 1,400 full-time employees from large companies in the US, tracked the frequency of AI use and its effects on cognition.

Approximately 14% of respondents reported a form of "mental fatigue" after intense conversations with AI systems, with symptoms such as difficulty concentrating, slower decision-making, and headaches. The study also states that the phenomenon may become more widespread as companies require employees to build and supervise more AI agents and juggle more tools.

For organizations adopting AI at scale, the signal is important. Productivity does not depend solely on access to tools, but also on the design of workflows, the number of systems people need to coordinate, and the cognitive cost of this constant oversight.

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