An AI model is trained directly on living human tissue

Outer Biosciences is developing a system that preserves living human skin tissue in the laboratory for up to a month and uses the resulting data to train an AI model. The tissue comes from surgical interventions and is provided by biobanks, with documented consent and without direct identifiers of the donors.

The model predicts which untested substances could influence a specific function of the skin. The substances are applied to the living tissue, and the results of the experiments are fed back into the system, improving subsequent predictions. The extended duration of maintenance allows for the observation of processes such as barrier restoration, pigmentation, or collagen remodeling.

The closed loop between prediction and testing generates biological data that do not exist on the internet and that can become a competitive advantage that is difficult to replicate. The company claims it has reduced the discovery rate from a few results in 18 months to a new candidate approximately every six weeks.

Outer Biosciences has raised approximately $23 million and intends to license the discovered ingredients to companies in the cosmetics or pharmaceutical industry, instead of building its own consumer brand.

AI data centers are moving outside major European cities

European developers of dedicated artificial intelligence data centers are increasingly looking for locations far from major cities. The main criteria are faster access to electrical networks, lower costs for energy and land, as well as the availability of sufficiently large areas for large-scale computing campuses.

According to JLL data, projects scheduled for the 2026-2028 period are, on average, 175 kilometers away from a major hub, compared to 46 kilometers for centers delivered between 2022 and 2025. Greenfield projects represent 39% of the future European portfolio, while locations within cities are expected to decrease from 13% to 5%.

This shift reflects the specific infrastructure requirements of AI, which consumes large amounts of energy and water for cooling. Secondary regions can attract investments and jobs, but projects may generate local opposition due to pressure on resources and habitats. For operators, the location of infrastructure becomes an energy and operational decision, not just one related to proximity to users.

JLL estimates that the four largest global cloud providers will spend $725 billion by 2026, 77% more than in 2025. By 2030, AI workloads could account for approximately half of the global capacity of data centers.

ChatGPT introduces ads in Europe

OpenAI is expanding the display of ads in ChatGPT in Europe, after testing the format in other markets. Ads will appear for users of the free versions and will be presented separately from the responses generated by the chatbot, in a distinctly marked area.

The company claims that the ads will not influence ChatGPT's responses and that users' conversations will not be sold to advertising agencies. The introduction of advertising provides OpenAI with a new source of revenue to support high operating costs, while maintaining free access to the service.

For the European digital market, the change brings the economic model of AI assistants closer to that of search engines and social platforms. The clear separation between responses and sponsored content, transparency regarding targeting, and compliance with European data protection rules will influence user trust and the adoption of the format.

Trust in AI remains low, and the companies behind it are viewed even more critically

A new study shows that the public continues to view artificial intelligence with reservations, and the level of trust in the companies developing these systems is even lower. The results indicate a clear difference between interest in the utility of the technology and the perception of those who build and manage it.

Concerns are related to how data is used, the lack of transparency, errors in the systems, and the impact of AI on jobs and public information. The rapid pace of launches has not been accompanied, in the perception of many users, by sufficient explanations regarding accountability, safety, and control mechanisms.

For companies implementing AI, trust becomes a product condition, not just a communication concern. Clear data policies, independent assessments, the possibility of human intervention, and explaining limitations can matter as much as model performance in the adoption decision.

Virtual AI company platforms try to manage emotional dependency

EVA AI, a virtual company platform, has hired a psychotherapist presented as the first specialist in relationships with AI partners. The initiative arises against the backdrop of discussions about emotional attachment to chatbots and the situations in which digital interactions begin to replace relationships and activities in real life.

Consulted specialists draw attention to the fact that these systems are designed to respond constantly and to affirm the user, which can make the relationship seem simpler than a human one. At the same time, the founder of an application dedicated exclusively to human meetings believes that interest in AI relationships is beginning to be replaced by the need for real connections.

For the consumer AI products industry, the subject highlights the need for design limits, clear messages about the artificial nature of the interlocutor, and support mechanisms for vulnerable users. An engaging experience can increase retention but can also amplify risks when the product simulates intimacy without the responsibilities of a human relationship.

An AI application created in Egypt helps the visually impaired understand their surroundings

Egyptian entrepreneur Karen Morad, who lost her sight, has developed an AI-based application to help visually impaired users interpret the world around them more easily. The solution uses the phone's camera and responds in real-time to questions about objects, texts, and visual situations.

The application supports 20 languages, including Arabic, an option that the founder identified as being insufficiently covered by existing tools. The product's construction thus starts from the direct experience of the user and from a local need that was not fully addressed by global services.

The project shows how multimodal models can transform the smartphone into a permanently available accessibility interface. The practical value, however, depends on the accuracy of the descriptions, the speed of response, and linguistic adaptation, essential elements when the generated information can influence mobility and daily decisions.

The humanoid robot Lightning breaks the human world record in the 100 meters

A Chinese humanoid robot named Lightning has completed the 100-meter trial in a time better than the world record set by a human athlete. The demonstration marks a significant progress in the speed of bipedal robots and in the dynamic control necessary to maintain balance at high accelerations.

The performance should be viewed in the context of a controlled technical trial, different from the conditions of an athletic competition. However, it shows how quickly mechanical components, actuation systems, and algorithms that coordinate movement in real-time are evolving.

For the robotics industry, speed is just one of the steps towards useful systems in real environments. The next challenges lie in autonomy, safety, energy efficiency, and the ability to work in unpredictable spaces. Sporting advances provide visibility for the technology, but commercial applications will be evaluated based on reliability and repeatable utility.

Divine brings back the Vine format to a social network without AI content

Divine, an application inspired by Vine and supported by Jack Dorsey, brings back six-second short videos to a platform built around human-created content. The application includes a large archive of Vine clips and allows the publication of new materials, created directly in the application.

The platform prohibits AI-generated content and uses technical checks for authenticity. Users can choose the type of algorithm used for the feed or switch to chronological display, an approach that seeks to reduce dependence on recommendations optimized solely for engagement.

Divine tests whether nostalgia and user control can support a social product in a market dominated by TikTok, Instagram, and automatically generated content. Its model puts the provenance of materials and open architecture in the foreground, but success will depend on forming an active community and an economic model compatible with the promise of a less intrusive experience.

Unitree goes public and brings humanoid robots to the attention of investors

Chinese manufacturer Unitree Robotics is preparing for its debut on the Shanghai stock exchange, after an initial public offering that raised over $900 million. The listing would make the company the first humanoid robot manufacturer traded on the continental market in China.

The offering, which targets approximately 10% of the company, indicates a valuation of around $9 billion and has been oversubscribed by more than 8,000 times. Unitree reported revenues of over $250 million in the previous year and, unlike many companies in this segment, operates profitably.

Investor interest confirms the accelerated pace of the market, but the gap between demonstrations and general applications remains significant. Robots can run, jump, or dance in controlled environments, but use in factories and homes requires autonomy, adaptation, and safety. The integration of agency AI can expand capabilities without eliminating the difficulty of interacting with the physical world.

The Galaxy S27 may receive a major design change

Initial unofficial information about the Samsung Galaxy S27 indicates a possible significant redesign, with visual elements close to the direction adopted by Apple. Images and renderings that have emerged before the launch suggest changes to the camera assembly and the overall shape of the device.

The information comes from data leaks and should be treated as preliminary until the official presentation. Samsung has gradually changed the design of the Galaxy series in recent years, and a more extensive revision could aim at both differentiating the new generation and standardizing the portfolio.

For the premium smartphone market, design remains an important criterion in a context where performance differences are harder to perceive. A visible change can support commercial positioning, but the product's value will also depend on hardware-software integration, cameras, battery life, and the AI features offered to users.

Applications for racing pigeons show the utility of an offline network

Applications dedicated to managing racing pigeons have emerged, an old communication technology that continues to have advantages in certain contexts. Digital solutions help track birds, organize routes, and monitor activity, combining a biological transport system with modern management tools.

Pigeons do not rely on mobile networks, electricity, or internet infrastructure to return to their point of origin. This autonomy explains why the method remains relevant for enthusiasts and for situations where conventional communications may be unavailable.

The example illustrates a useful idea for designing infrastructure: the newest technology is not automatically the most robust for every scenario. Resilient systems sometimes combine simple, distributed, and network-independent components with software for coordination, traceability, and analysis.

A compact AI agent reproduces research better than much larger models

Inherently, a London-based lab founded by former Google DeepMind researchers claims that its agent Faraday has outperformed frontier models from Anthropic and OpenAI in a specific task: independently reproducing results from published scientific papers, without receiving the answer in advance.

Faraday uses Qwen 3.6, a model with 27 billion parameters, much smaller than the systems it was compared to. The company aims not only for accuracy but also what it calls "taste for research": selecting relevant experiments and designing them.

Training is based on rewarding good results through reinforcement learning.

The outcome suggests that the architecture of the agent, the objectives, and the workflow may matter more than the size of the base model. Inherently, it wants to build an AI collaborator capable of contributing to scientific discoveries and uses external tools, including Codex, instead of reconstructing every component.

The company recently emerged from stealth mode with a $50 million seed round and intends to expand its team from about 12 to 20-25 people by the end of the year.

Nvidia shows why the system around the AI model becomes decisive

A study published by Nvidia shows that the performance of an agent in long-duration tasks heavily depends on the "harness", that is, the software layer that surrounds the model. This includes available tools, memory, execution rules, and components that coordinate the agent's steps.

By optimizing memory management and adding a monitoring component, researchers achieved a 100% score for Claude Opus 5 on ARC-AGI-3. The benchmark contains two-dimensional interactive games without instructions, in which the system must discover the rules on its own and find a winning strategy.

The conclusion is relevant for teams building AI products: choosing the model is not enough. Orchestration, context persistence, result verification, and control mechanisms can radically change performance. The competitive advantage thus shifts from access to a powerful model to the quality of the entire operational system.

OpenAI regains ground among business clients

Data published by the expense management platform Ramp indicates that OpenAI is gaining ground against Anthropic among companies in the United States. In the absence of detailed public financial statements from the two companies, data on organizational payments provides an indirect indicator of commercial adoption.

The evolution suggests that the enterprise market remains fluid, and a provider's advantage can change rapidly with the launch of new models, features, and integrations. Companies may use multiple providers simultaneously or shift workloads between platforms based on cost, security, and performance.

For IT teams, tighter competition can bring better prices and more advanced capabilities, but it also reinforces the need for a flexible architecture. Evaluating providers must include data portability, governance, model availability, and the cost of switching, not just the results of a snapshot benchmark.

ChatGPT can send messages through a new Apple Messages plug-in

OpenAI has launched a plug-in that connects ChatGPT to the Apple Messages inbox. Users can ask the chatbot to sort, analyze, and edit conversations, as well as prepare or send messages based on existing context.

The integration also works with Codex and ChatGPT Work, extending usage from personal communication to professional workflows. An example presented by the company shows how the assistant can analyze messages received in a day and propose follow-up responses for contacts.

Access to an AI agent in private conversations can save time but introduces important requirements regarding permissions, action confirmations, and data protection. In the business environment, organizations must establish which conversations can be processed, who can authorize sending, and how logs of actions taken on behalf of the user are maintained.

Google allows publishers to add a button to become preferred sources

Google allows publishers to include a button on their own sites through which readers can select them as "preferred sources". The choice can increase the visibility of the publication in Google Search, Discover, and Google News, at a time when AI-generated responses reduce traffic sent to original pages.

The feature extends the Preferred Sources option, previously introduced in Google's AI experiences, including AI Mode and AI Overviews, after having been available in Top Stories. Publishers thus receive a direct tool to transform loyal readers into a distribution signal in the Google ecosystem.

The measure does not fully resolve the traffic decline caused by AI summaries, but it can strengthen the relationship between publishers and recurring audiences. For editorial products, brand identity, subscription, and direct channels become more important, as simple optimization for search offers less control over access to the audience.

AI-supported hackers target vulnerable water supply systems

U.S. security agencies warn that attackers are compromising Siemens devices used in critical infrastructures. The targeted programmable controllers from the S7 series are used for automating processes in water supply, energy, production, and agriculture.

CISA, FBI, and NSA state that attackers are also using AI tools to identify targets, prepare attacks, and exploit poorly protected systems. Industrial equipment can remain operational for long periods, and exposure to the internet, default passwords, and delayed updates increase the attack surface.

For infrastructure operators, the incident confirms the need for asset inventory, segmentation between IT and operational networks, updating controllers, and continuous monitoring. In the case of physical systems, a cyber attack can directly affect public services, and using AI reduces the cost and time required for attackers to conduct reconnaissance.

Google expands AI tools for study in Search and Gemini

Google has announced new educational features in Search and Gemini, including AI-generated interactive visualizations, 3D simulations, a hub dedicated to students, and adaptive verification tests. The tools aim to transform the assistant into a learning space, not just a source of quick answers.

The features can adapt exercises to the user's level and present complex concepts through visual representations. The launch intensifies competition with OpenAI and specialized educational platforms that already offer homework assistance, explanations, and personalized practice.

For education, the benefit depends on how the product supports the reasoning process. Progressive feedback, source verification, and understanding assessment are more valuable than generating the final outcome. Institutions will also need clear rules regarding AI use, academic integrity, and student data protection.

Biological data, not just models, can accelerate cancer research

The startup Vivodyne claims that the main limitation of AI in discovering cancer treatments is the lack of causal biological data obtained from human tissue. The company has built HIVE, modular robotic laboratories that can grow 20 types of tissue and can automatically manage and monitor experiments.

Current models often rely on data from animal testing, individual cells, or isolated proteins. These sources do not fully replicate the behavior of living human tissue, which can lead to promising results in the lab that do not confirm in clinical studies.

Vivodyne's approach automatically connects physical automation, experimental measurement, and AI analysis into a learning cycle. For biotechnology, infrastructure capable of producing new and relevant data may be more important than a larger model. The value of the system lies in the quality of the experiments and their proximity to real human biology.

Warp launches a complete infrastructure for software factories with AI agents

Warp has introduced Warp Factories, a system designed for building and operating "software factories" based on AI agents. The concept organizes agents into a loop that covers traditional development stages, from defining the task and writing code to verification and delivery.

The product functions as an infrastructure layer for deploying agents and provides companies with a preconfigured working framework. The goal is to reduce the effort required to connect tools, permissions, and processes that transform a coding agent into a usable system by a team.

For software organizations, the promise is a more coherent automation than the isolated use of an assistant. Adoption, however, depends on the quality of controls, integration with code repositories, testing, and human approvals. An efficient software factory must produce verifiable and easily auditable results, not just more code.

Reddit tests an audio-video feed inspired by TikTok

Reddit is starting to test an experience that combines video and audio posts in selected communities. The platform wants to transform popular stories into an easy-to-follow or listen-to format in the background, without users needing to read the entire posts.

The test will seek to determine which types of content have sufficient relevance to be scaled up. The format brings Reddit closer to the vertical feeds popularized by TikTok, but uses conversations and community archives as sources for audio-video materials.

For Reddit, the change could increase the time spent in the app and make content accessible in more contexts. At the same time, transforming written posts raises questions about author attribution, moderation, and maintaining context. The product will need to balance rapid consumption with the specifics of the communities that differentiate the platform.

British cinemas are considering restrictions on Meta glasses

The UK Cinema Association is considering restricting the use of Meta smart glasses in cinemas due to piracy and unauthorized filming risks. The devices have built-in cameras and can discreetly record, and the indicator light can be easily covered.

The association estimates that copyright infringement costs the UK creative industries at least £500 million annually. Over 90% of illegal copies of newly released films come from recordings made in cinemas, and by 2025, cinema staff reported 121 incidents of illegal filming.

A general ban would, however, also affect legitimate uses. Some models can display custom subtitles and other useful features for people with visual or hearing impairments. The debate highlights the difficulty of establishing rules for wearable devices that combine accessibility with a hard-to-detect recording capability.

Robots could transform volcanic lava into building material

Researchers are exploring the use of robots for collecting and processing lava during volcanic eruptions. The cooled material could become an alternative to components used in concrete, leveraging a natural resource available in large quantities in certain regions.

The volcanic environment is too dangerous for direct human intervention, which is why autonomous systems would need to withstand extreme temperatures and unstable terrain. The project combines robotics, real-time monitoring, and materials science to control a process that is normally difficult to access.

If the method can be scaled, it could reduce dependence on the extraction of raw materials and the emissions associated with construction. Viability depends on safety, equipment costs, the chemical characteristics of the lava, and the ability to transform an unpredictable phenomenon into a stable industrial flow.

IBM connects cryogenic modules to scale quantum computing

IBM has demonstrated the connection of cryogenic modules, an important step for expanding quantum systems beyond the limits of a single dilution refrigerator.

Superconducting quantum computers operate at extremely low temperatures, and the space and wiring of a single cooling system limit the number of qubits.

The modular architecture allows the distribution of components across multiple units maintained in cryogenic conditions and their connection into a larger system. The approach aims to solve the physical challenges of scaling, not just to increase the performance of the quantum processor.

For the quantum industry, interconnecting modules can play a role similar to networks in classical data centers. Progress will depend on the fidelity of the connections, error control, and the capacity of the cooling infrastructure. The demonstration shows that the evolution of quantum computing is as much related to system engineering as it is to qubit design.

Preference for smartphones may reflect different cognitive styles

An analysis explores whether the choice between smartphone ecosystems can be associated with different ways of thinking and decision-making. The comparison starts from the popular opposition between a more structured approach and a more flexible one, applied to preferences for iPhone and Android.

The concept of "left brain" and "right brain" is a simplification, and buying behavior is influenced by numerous factors: price, familiarity, integration with other devices, customization options, and perceptions of security. The choice of platform cannot be reduced to a unique neurological trait.

For companies developing applications, the relevant differences are those of behavior and expectations, not psychological labels. User research, usage data, and testing across both ecosystems provide a more solid basis for product decisions than assumptions about the personality of device owners.

Japanese companies adopt AI more slowly than other markets

Companies in Japan are proceeding cautiously in adopting artificial intelligence, despite the economy facing labor shortages, an aging population, and persistent productivity issues. These conditions could favor automation, yet the use of AI in the business environment remains behind that of other developed markets.

Among the indicated explanations are risk aversion, a conservative organizational culture, and decision-making processes that require consensus. Companies are cautious regarding errors, data security, and reputational effects, which prolongs the period between testing a technology and implementing it on a large scale.

The slow pace may limit short-term productivity gains but also provides space for more controlled implementations. For software providers, the market demands solutions tailored to local processes, clear evidence of reliability, and integration with existing systems, not just access to the latest AI models.

Meta collects three times more data than Microsoft or Apple

A comparative analysis shows that Meta collects approximately three times more categories of data about users than Microsoft or Apple. The difference reflects the company's economic model, heavily dependent on advertising and personalizing the experience on its platforms.

The data may include identification information, online activity, interactions, location, and elements used for profiling. The volume does not automatically indicate a violation, but it increases the complexity of consent and the difficulty for users to understand what information is associated with their account.

For companies and developers, the comparison underscores the importance of the data minimization principle. Collection should be linked to a clear product need, and retention and sharing must be documented. In Europe, transparency and user control are essential for both compliance and trust in digital services.

AI changes the core competencies of young programmers

Generative tools are transforming how early-career programmers learn and work. An increasing portion of code can be produced automatically, and the focus is shifting from memorizing syntax to defining problems, evaluating solutions, and understanding architecture.

This change can accelerate learning and prototyping but risks reducing exposure to the exercises that form technical intuition. If a developer accepts code without understanding its behavior, security errors and maintenance issues can reach the product more quickly.

For software teams, training juniors must include responsible use of AI, code review, debugging, and explaining decisions. Professional value is shifting towards reasoning, verification, and collaboration with automated systems, but the fundamentals of programming remain necessary to control outcomes and intervene when the tool makes mistakes.

A Spanish nanomaterial cools surfaces without electricity

Researchers in Spain have developed a nanomaterial capable of reducing the temperature of surfaces by up to 12.9 degrees Celsius without electricity consumption. The material uses radiative cooling, reflecting a large portion of solar energy and eliminating heat in the form of infrared radiation.

The solution can be applied to buildings and other surfaces exposed to the sun. Unlike air conditioning, the process is passive and does not require compressors, refrigerants, or continuous power, making it relevant in regions with high temperatures and limited energy infrastructure.

On a large scale, the material could reduce consumption for cooling buildings and pressure on networks during peak periods. The next steps involve durability, production costs, and integration into commercial materials, conditions necessary for laboratory performance to become an infrastructure solution.

Google offers students a free year of Gemini AI Pro

Google is offering eligible students free access for a year to the Gemini AI Pro plan. The initiative aims to introduce the company's premium tools into academic activities and attract a generation of users who are now shaping their digital study and work flows.

The plan includes access to more advanced Gemini models and features for research, writing, organizing, and analysis. The offer fits into a broader competition among AI providers for integrating assistants into education and transforming them into daily-use tools.

For universities and students, extended access can increase productivity but imposes rules regarding citation, confidentiality, and individual assessment. The free offer reduces the financial barrier in the short term, but users must understand the plan's conditions and the costs that may arise after the promotional period ends.

AI has structurally changed the memory industry, says Micron CEO

The CEO of Micron states that the demand generated by artificial intelligence has fundamentally changed the memory industry. AI-dedicated data centers use large volumes of high-performance memory to power accelerators and rapidly move data between system components.

The growing demand for HBM and other advanced products is changing investment plans, production capacity, and the relationship between memory manufacturers and cloud companies. Memory is no longer treated just as a standardized component but as a critical element for the performance of the entire AI infrastructure.

For the hardware market, this can support prices and investments but can also create supply bottlenecks. Companies building infrastructure need to plan long-term purchases and track not only the availability of processors but also memory capacity, energy consumption, and system integration.

Economists warn that the AI stock rally may suffer a correction

The rapid growth of stocks associated with artificial intelligence has fueled discussions about a possible market correction. Economists observe that high valuations include very high expectations regarding future revenues, productivity, and the pace of technology adoption.

Massive investments in chips, data centers, and models can create long-term value, but monetization does not advance uniformly across all sectors. A slowdown in demand, energy costs, or financial results below estimates could lead to a reevaluation of companies heavily exposed to the AI theme.

A stock market correction would not automatically mean that the technology has no economic impact. For companies, the important difference is between infrastructure and products that generate measurable results and projects primarily supported by expectations. Investment discipline and return on investment evaluation become more important as the market matures.

Residents of Michigan sue a data center for noise pollution

Several residents in Michigan have filed a lawsuit against the operator of a data center, claiming the constant noise produced by the facility. Cooling systems, fans, and generators can operate continuously, and the low-frequency sound affects nearby communities.

The conflict reflects the rapid expansion of digital infrastructure in areas that have not been prepared for industrial installations of this size. Residents claim that the noise affects their sleep and quality of life, while developers must comply with local limits and permit conditions.

For the data center industry, community acceptance becomes a component of project risk. Acoustic studies, equipment placement, noise barriers, and dialogue before construction can influence the timeline and costs. The impact of AI infrastructure is not limited to energy and water but also includes direct effects on the areas where it is built.

A teenager drops a test case regarding social media addiction

A teenager from New Jersey has dropped the case that was to become the next test case against Meta, YouTube, and Snap for the alleged addictive effects of the platforms. The withdrawal changes the litigation calendar through which plaintiffs are trying to establish the companies' liability for the effects of their products on young people.

The cases argue that features such as infinite feeds, notifications, and recommendation systems are designed to maximize time spent in the app and can affect mental health. Companies contest the allegations and invoke both the lack of a simple causal link and the safety measures introduced for minor users.

Even if this case does not continue, the legal pressure on the platforms remains high. Future outcomes may influence product design, age verification, and transparency obligations. For the industry, the safety of adolescents becomes a requirement that must be integrated into architecture, not just treated through optional settings.

AI advances trigger a crisis of identity in mathematics

The new Astra model from OpenAI has presented results for ten important problems in mathematics and theoretical computer science, including in quantum game theory and sphere packing in higher dimensions. Consulted researchers considered the contributions solid enough that achieving any of them would have been remarkable for a human mathematician.

The evidence has been published and formalized with Lean, a language that allows logical verification of proofs. However, questions remain about the number of attempts, the level of guidance provided by human experts, and how previous works were credited. Success in a set of problems does not demonstrate a general capability across all branches of mathematics.

The impact goes beyond technical performance. Mathematicians discuss the role of the researcher, funding for the field, and the risk that automation will solve problems without opening new directions. At the same time, tools can democratize access to advanced research, provided that users can verify results and understand the limitations of the models.

Slack launches channels where teams program together with AI agents

Slack introduces Slack Code, dedicated channels for development projects where teams can work directly with AI agents. A user can tag an agent like Claude or Devin, and the system creates a separate channel for task resolution, without moving the conversation to other tools.

The channel displays code changes, HTML previews, and feedback provided by colleagues before delivery. Team members can follow the discussion, audit differences, and approve the result, and the channel is automatically archived after the project is completed, keeping a log of actions.

The feature is available on all Slack plans and integrates with agents from the marketplace, including Claude Code, Devin, Vercel Agent, and GitHub Copilot. For teams, moving agentic coding into a collaborative space can improve transparency but requires clear policies for access to repositories, review, and approval responsibility.

Google presents how foreign app stores could look in Google Play

Google has presented a prototype for integrating foreign app stores into Google Play, as part of changes imposed following litigation with Epic Games. The experience would allow users to discover and install competing stores within the Google ecosystem.

The measure aims to open up the distribution of Android applications and reduce the structural advantages of the Play Store. Implementation must provide rival stores with real access to users while maintaining the necessary warnings and checks for security.

For developers, a more open distribution could mean additional payment options, different commercial terms, and access to alternative channels. For Google, the challenge is to meet competition requirements without creating frictions that make foreign options unusable and without transferring risks to users that they cannot evaluate.

Apple Music will display labels for content created with AI

Apple Music intends to display, during this year, visible labels for materials in which AI has been used to a significant extent. These will transform into visible information for the public the Transparency Tags metadata introduced by the company in March.

Content providers will need to mark tracks generated by AI platforms or materials in which a significant part of the recording was created with such tools. The system can also cover composition, album graphics, or video materials, depending on the type of automated contribution.

Labeling can provide listeners with more context and support transparency in music distribution. However, efficiency will depend on the enforcement rules and the verification of statements submitted by artists, record labels, and distributors. Without a coherent mechanism, voluntary metadata may remain incomplete.

Adobe makes AI generators for music, voice, and effects available to the public

Adobe has launched to the general public the Generate Music, Generate Speech, and Generate Sound Effects tools from Firefly, after beta testing periods. Users can create audio elements from text instructions and integrate them into production flows within the Adobe ecosystem.

The features cover three distinct needs: background music, synthesized voice, and sound effects tailored to a scene. General availability indicates a shift from experiment to commercial tool, with a focus on use in video materials, campaigns, and creative projects.

For professionals, audio generation can reduce the time needed for prototyping and can complement traditional libraries. Adoption depends on control over style, the quality of the output, and the clarity of usage rights. Integration into existing applications can be a more important advantage than isolated models, as it preserves editing and approval in the same flow.

OpenAI launches ChatGPT for teens

OpenAI is launching ChatGPT for Teens, a version aimed at users aged 13 to 17. The product includes additional protections for conversations about suicide, self-harm, and romantic or sexual themes, amid concerns about adolescents' emotional dependency on chatbots.

The version also offers support for homework and study, designed to help students learn instead of generating essays or final answers directly. The company says it is trying to adapt the experience to the developmental stage of adolescents, without treating them as small children or adults.

The launch responds to documented risks in previous tests, where chatbots could provide dangerous details even after displaying warnings. A Common Sense Media study from 2025 would show that over 70% of American teenagers use chatbots for companionship, and half regularly turn to such services.

Age-specific protections may reduce exposure to harmful content, but their effectiveness will depend on the correct detection of risk situations, age verification, and directing users to human help when the conversation exceeds the role of a digital assistant.

Training AI on copyrighted books remains an open legal issue

AI models are trained on very large volumes of published texts, including books protected by copyright, often without the explicit consent of the writers. The legality of the practice does not have a single answer, as courts analyze separately the use of works for training, how copies were obtained, and the results produced by the model.

Companies invoke the fair use doctrine and argue that the process is transformative, while authors claim unauthorized copying and the economic impact on their work. Some rulings have differentiated between the training itself and the use of collections from pirated sources, complicating the establishment of a general precedent.

For model developers and applications, the provenance of data becomes a central legal and operational risk. Licensing, documenting datasets, and mechanisms through which creators can exercise their rights can influence the cost and sustainability of AI products.

As processes continue, the industry will operate within a fragmented framework, where the same practice may be evaluated differently depending on jurisdiction, the source of materials, and the concrete behavior of the system.

More than a third of new web pages show signs of AI authorship

A Pew Research study shows that more than a third of pages published after the launch of ChatGPT show signs of having been written or substantially modified with AI. The analysis confirms the extent to which generative tools have entered online content production.

Researchers analyzed nearly half a million pages in English from the Common Crawl archive, covering approximately five years, including the period before the launch of ChatGPT in November 2022. Open Pangram technology was used for identification.

The increase in automatically generated content occurs in parallel with the expansion of traffic generated by bots. The result is a web where automated systems publish materials that other systems index, summarize, or use for training, with the risk of degrading quality and multiplying errors.

For publishers and companies, differentiation no longer comes just from the speed of publication. Verifiable sources, expertise, direct experience, and editorial processes become necessary elements for materials to remain useful in an increasingly large volume of automated content.

Artificial intelligence may increase short-term inflation

Artificial intelligence may exert upward pressure on inflation in the short and medium term, says Petra Tschudin, a member of the governing body of the Swiss National Bank. The overall effect remains uncertain, as technology can influence prices in both directions.

Investments are being redirected towards AI infrastructure, which can create adjustments and difficulties in other sectors. High demand for chips, energy, and data centers can generate capacity shortages and price increases before productivity benefits spread throughout the economy.

In the long term, higher productivity may reduce the cost of goods and services. Tschudin emphasizes, however, that inflation is calculated annually, and for a structural deflationary effect, price decreases would need to be repeated. Increases in productivity do not automatically lead an economy towards deflation.

For companies, the transition to AI may include a period in which implementation and infrastructure costs arise before savings. Planning must take into account both operational returns and price volatility for critical components.

Repetitive AI content for children may amplify cognitive and behavioral risks

A study by the Global Network on Extremism and Technology warns that repetitive, low-quality, and sometimes violent AI videos distributed massively to children may contribute to cognitive decline and normalize aggressive behaviors. The analysis pays attention to animated clips with fruits, which are very present in feeds aimed at minors.

The so-called "AI slop" is cheap and produced in large volumes, with minimal human intervention, to maximize views, clicks, and advertising revenue. A single creator can publish hundreds of short materials on multiple platforms, taking advantage of systems that reward frequency and engagement.

Researchers warn that the format can hide or normalize violent and extremist messages in seemingly innocuous content. Repetition, rapid pace, and lack of a coherent narrative can affect children's attention and understanding, especially when automated recommendations continuously deliver similar materials.

For platforms, the problem cannot be solved just by detecting an image or a forbidden word. Context evaluation, limits for the distribution of mass-produced content, parental controls, and accountability for how algorithms recommend materials to minor users are necessary.

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