IT News Review by Control F5 Software: AI chatbots can influence political opinions using inaccurate information

Study: AI chatbots can influence political opinions using inaccurate information

A study cited by NBC News shows that AI chatbots can change users' political opinions when they provide biased or inaccurate information about candidates and public policies. Controlled experiments demonstrated that participants exposed to distorted responses adjusted their preferences, even when the information was not factually correct.

For software producers, social platforms, and AI governance teams, the results underscore the need for strict moderation policies for politically generated content by models, as well as transparency regarding information sources. In the context of elections, it is essential that automated systems are limited in recommendations and clearly marked as tools that can make mistakes.

—

ChatGPT slows down, but its competition accelerates

A new report on the generative AI application market shows that the growth rate of ChatGPT's user base has slowed in recent months, even though the application remains a market leader in downloads and monthly active users. Meanwhile, rival models like Google Gemini and DeepSeek are gaining ground, especially in markets where they are better integrated into local ecosystems or existing products.

For IT and business professionals, the message is that the AI space is beginning to behave like a mature market, not a stable monopoly: differentiation through product quality, integration into workflows, and regional positioning matter at least as much as "first mover" advantage. Adoption strategies should consider a portfolio of models, not exclusively betting on a single provider.

—

OpenAI halts app recommendations that look like ads

OpenAI has disabled a new type of app recommendations in ChatGPT after users reported that they looked too much like ads inserted into the interface. The suggestions appeared as "cards" for brands like Target or Peloton, and the company later clarified that they are not paid ads and that no commercial ad tests are currently running in ChatGPT.

The incident shows how sensitive the monetization area is for conversational AI products: the difference between a contextual recommendation and a format perceived as intrusive advertising is very small at the UX level. For product and marketing teams, the lesson is that monetization tests must be communicated transparently, with a clear separation between the model's response and content with commercial potential.

—

Meta delays mixed reality glasses until 2027

Internal sources cited by the press show that Meta has postponed the launch of its Phoenix line mixed reality glasses from the initial 2026 window to the first half of 2027. The device was intended to be a direct competitor to Apple Vision Pro and future mixed reality headsets on the market, but the company is reevaluating both hardware costs and consumer appetite for such a premium product.

For the software ecosystem and companies planning MR applications, this delay shifts the horizon in which this category can become mainstream. Meanwhile, Meta seems to prioritize Ray-Ban glasses with AI features and conversational assistants, suggesting that, in the short term, "hands-free" voice and camera-based experiences will receive more attention than fully immersive mixed reality experiences.

—

Yoodli, an AI assistant for public speaking, reaches a valuation of $300 million

The startup Yoodli, founded by former Google employees, has raised a new round of funding that triples the company's valuation to approximately $300 million. The platform uses AI to analyze speeches, presentations, and conversations and provides granular feedback on structure, pace, clarity, or non-verbal language, with a clear emphasis on assisting people, not replacing them.

For HR, L&D, and leadership teams, such solutions mark an interesting direction: AI as a training tool for soft skills, including presentations, sales, or managerial communication. Integrating such a product into internal training programs can increase scalability and consistency of feedback, without eliminating the role of human trainers, but rather complementing it.

—

Meta acquires AI device startup Limitless

Meta has acquired Limitless, the company behind a "necklace" wearable device and personal information capture and retrieval software. As a result, Limitless has halted hardware sales and is changing its business model, with part of the team working on Meta's wearable and AI products. The solution was known for its conversation recording and summarization feature, using generative models to transform meetings into notes and actions.

This move strengthens Meta's direction towards wearable devices with integrated AI, where capturing context from the real world (audio, video, location) becomes a competitive advantage. For software companies, this suggests a future where some applications will be designed directly for "ambient" interfaces, not just for screens: automatic meeting summaries, AI-assisted personal journaling, recalling context from projects.

—

The New York Times sues Perplexity for copyright infringement

The New York Times has filed a lawsuit against Perplexity, accusing the company of using its editorial materials for training and responses of the AI search engine without a proper license. The complaint mentions cases where Perplexity allegedly reproduced nearly entire paywalled articles or generated new materials based on the newspaper's content without clear citation.

This litigation adds to other lawsuits filed by publishers against AI companies and intensifies discussions about how models can be trained on copyright-protected content. For software providers and product teams integrating large language models, the legal risk becomes an important selection criterion, favoring solutions with explicit licensing agreements or the ability to limit data sources.

—

Meta signs commercial agreements to bring real-time news into Meta AI

Meta has announced the conclusion of data licensing agreements with several news publishers, allowing it to display real-time content in Meta AI, its conversational assistant. Instead of relying solely on web crawling, the company pays for structured access to news feeds, which can then be summarized or contextualized by the model.

For the industry, this model shows what a "new normal" between AI and media could look like: commercial agreements, clean feeds, and visible citations, instead of the implicit use of content. For software and technology companies, access to such licensed feeds can become a differentiator for their own knowledge management or assistance products.

—

Anthropic CEO talks about the "AI bubble" and risk appetite in the industry

In a recent interview, Dario Amodei, CEO of Anthropic, acknowledged that there are elements of a bubble in current investments in AI, but emphasized that his firm is trying to maintain a balance between growth and caution. He implicitly criticized some competitors' tendency to launch "YOLO" products without sufficiently robust security and testing processes, insisting on the need for serious technical evaluations before exposing very powerful models to the public.

The message is relevant for CTOs and product leaders: the pressure to keep up with the pace of launches cannot substitute for technical governance. Implementing internal safety review processes, red teaming, and abuse scenario testing should be part of the culture of any organization integrating frontier models into critical or market-oriented products.

—

Meta centralizes support for Facebook and Instagram and tests an AI assistant

Meta has launched a unified support hub for Facebook and Instagram, which gathers reporting issues and account recovery options in one place. The company explicitly acknowledges that the support service "has not always met expectations" and announces the testing of an AI assistant for instant help, from account recovery to updating security settings.

From a technical perspective, this move suggests a wave of automation in the AI-based customer support area, including for sensitive issues such as account compromise. For corporate users and creators, a clearer recovery flow and faster support can reduce reputational and operational risks related to access to pages and accounts with large audiences.

—

Meta cuts up to 30% from the metaverse budget

According to insider information, Meta is considering cutting the budget for metaverse initiatives by up to 30% and redirecting a significant portion of these resources to AI-related projects and smart glasses. The cuts could affect both VR/AR hardware development and certain software teams dedicated to immersive experiences.

For the digital ecosystem, this confirms the pivot of major platforms from the metaverse to generative AI and intelligent assistants. Companies that have invested in products exclusively aimed at VR environments need to reassess their roadmap, while AI-augmented productivity and collaboration solutions receive additional validation signals.

—

Gemini, the most searched term on Google in 2025

"Gemini" was the fastest-growing term in Google Search globally in 2025, according to the official "Year in Search" report. The interest reflects the successive launches of Gemini 2 and 3 models, integration into products like Chrome, Android, and Workspace, but also direct competition with ChatGPT in the conversational assistant space.

For business and IT, this trend confirms not only the appetite for AI in general but also the migration of attention towards products directly integrated into existing ecosystems (browser, phone, productivity suites). Strategies for integrating AI into internal applications can benefit from this behavior: users adapt more easily to "embedded" AI than to completely separate tools.

—

Google tests unifying AI Overviews with AI mode in Search

Google is experimenting with an interface in which AI Overviews – synthetic responses from the results page – are more closely integrated with "AI Mode", the conversational mode. In certain tests, users see a continuous flow: an initial automatic synthesis above classic results, followed by the possibility to continue the discussion in real-time with an AI agent, maintaining the initial context.

This design confirms the direction of transforming search engines into hybrid search + conversation interfaces. For companies, the impact is twofold: on one hand, SEO optics need to be extended towards "optimization for AI responses", not just for links; on the other hand, it becomes easier to imagine internal knowledge flows based on the same type of experience, across their own documentation.

—

Auracast: strong standard, weak marketing

Auracast, the Bluetooth technology that allows "audio broadcasting" to multiple devices – from headphones to hearing aids – has support from giants like JBL, LG, Samsung, and Google, but is promoted inconsistently and is hard for users to understand. The heterogeneous branding and lack of unified communication make a standard with great practical benefits seem like an obscure feature in menus.

From a technical perspective, Auracast can be relevant for conference rooms, public spaces, waiting rooms, or offices, where audio can be delivered discreetly to multiple users simultaneously. In the software ecosystem, opportunities for integration with event applications, real-time translation systems, or accessibility solutions for people with hearing impairments arise.

—

iOS 26.2 brings new features to CarPlay

An upcoming iOS update (26.2) will bring two new features for CarPlay: the ability to "unpin" messages to reduce clutter in the interface, and an expanded set of widgets that provide more real-time information, inspired by the experience from iOS 18.

For auto app developers and head unit manufacturers, the change indicates a continued focus on simplicity, glanceability, and control over notifications. Integrating a large volume of data – from navigation to media and messaging – must be done without overwhelming the driver, which will influence both UX design and the APIs used by third-party applications.

—

NTT Data CEO: the AI bubble will be short-lived

NTT Data leadership believes that the current "bubble" of investments in AI will be relatively short-lived, as companies will quickly demand concrete results and measurable ROI. In an interview, he emphasizes that solutions that address specific business problems and integrate into processes will survive, not just powerful general models without clear use cases.

For enterprise clients and software providers, the message reinforces the idea of "pragmatic AI": vertically-oriented products, POCs with clear metrics, and a focus on scaling projects that demonstrate visible effects (efficiency, cost reduction, revenue growth).

—

EU investigates Meta's AI access policy in WhatsApp

The European Commission has opened an antitrust investigation into Meta's decision to block rival chatbots like ChatGPT and Copilot in WhatsApp, imposing only Meta AI on users in certain scenarios. Authorities will determine whether this policy abuses the platform's dominant position and restricts competition in the conversational AI assistant market.

For companies building services over messaging, the case is important: if WhatsApp becomes a closed space for third-party AI, distribution and integration strategies need to be recalibrated, and European rules may set a precedent that influences how gatekeeper platforms integrate their own AI models.

—

Chart shows why AI could be "bigger than COVID"

A Sky News analysis, based on data from research firm Metr, shows that the "capability" of AI models for software development tasks has doubled approximately every seven months over the past few years. Marc Warner, CEO of the British company Faculty, argues that this type of exponential growth could have a deeper and more lasting impact than the COVID pandemic, if the trend continues for at least another five years.

For the industry, the key message is not the comparison itself, but the idea of preparing for a wave of rapid changes: increasingly capable models, decreasing inference costs, and growing pressure for automating cognitive tasks. Planning for infrastructure, skills, and governance must take into account a three-to-five-year horizon, not the classic ten-year cycles.

—

Samsung raises the stakes for foldables with Galaxy Z Trifold

Samsung is preparing a Galaxy Z Trifold device, a foldable phone-tablet with a screen that folds into three segments, opening up to nearly the size of a small laptop or large tablet. The article shows how the design targets heavy multitasking and content consumption users, but also raises questions about weight, mechanical complexity, and cost.

For developers, such extended form factors put pressure on responsive design and managing dynamic layouts: applications need to adapt fluidly from "phone" mode to two or three panels simultaneously, possibly with parallel sessions of the same application. This is an area where UI/UX optimization can become a real differentiator.

—

Memory crisis could raise prices for phones and laptops

Explosive demand for memory for AI data centers is starting to strain the supply chain for DRAM and NAND, which could lead to price increases for smartphones and PCs. Production is limited by the capacity of existing factories and the large investments needed for new facilities, and manufacturers prioritize enterprise clients with long-term contracts.

Companies equipping fleets of devices or producing hardware need to consider larger budgets and potential delays. For software, the impact comes indirectly, through longer upgrade cycles and the pressure to optimize applications for older hardware in certain customer segments.

—

Horizon launches first quantum computer for commercial use in Singapore

The startup Horizon Quantum Computing, backed by strategic investors, has activated the first quantum computer for commercial use in Singapore, in a center that will provide access to local and regional companies. The infrastructure is intended to be used for optimization, finance, logistics, and research, with a software layer that hides the complexity of quantum hardware.

Although the current performance of quantum systems still does not replace classical computing, early commercial access is a signal for developers who want to experiment with hybrid algorithms. Integration through APIs and SDK-type tools will allow technical teams to start exploring scenarios where "quantum-ready" can provide a competitive advantage in a few years.

—

Chrome can autofill forms using more data from Google Account and Wallet

Google has expanded autofill capabilities in Chrome on desktop, Android, and iOS, allowing the browser to use data from Google Account (name, email, addresses) and Google Wallet (documents, loyalty cards, travel details). On Android, the autofill suggestion interface above the keyboard has been redesigned to display more context and differentiate similar entries more easily.

For users, filling out travel or booking forms becomes faster, but it also increases the importance of careful permission management. For developers and e-commerce teams, form design needs to be compatible with these autofill flows, reducing friction in the funnel and avoiding overly restrictive validations that block pre-filled data.

—

Windows receives "out-of-the-box" MIDI 2.0 support

Microsoft will add MIDI 2.0 support in Windows in the coming months, after Windows MIDI services have reached the Dev and Beta channels of Windows Insider. The new generation of MIDI allows for higher resolution for expressiveness, bidirectional communication between devices, and reduces reliance on third-party drivers. macOS and Android have already adopted MIDI 2.0 in 2021-2022, but the ecosystem of compatible hardware is still in its infancy.

For audio software developers and music equipment manufacturers, native support in Windows will simplify integration and encourage the emergence of more expressive applications and instruments. In the content creation area, this could mean more sophisticated plugins and DAWs, including the integration of generative AI with fine control over performance.

—

Facebook and Instagram launch a dedicated hub for recovering stolen accounts

Meta has introduced a centralized support hub through which Facebook and Instagram users can report access issues and attempt to recover compromised accounts. The company admits that previous support has not met expectations and promises a clearer process, complemented by an AI assistant to guide users through verification and recovery steps.

For creators, brands, and companies that rely on these platforms for marketing, having a more structured support channel is essential. Beyond reactive measures, it remains important for IT and social media teams to apply strict security policies: 2FA, access control to pages, and monitoring of suspicious activity.

—

Gemini 3 Deep Think: advanced reasoning mode for Google AI Ultra subscribers

Google has launched the Gemini 3 Deep Think mode, a reasoning-focused variant of the Gemini 3 model, available to Google AI Ultra subscribers in the Gemini app. Deep Think is optimized for complex problems in mathematics, science, and logic and ranks first on the ARC-AGI-2 benchmark, dedicated to testing general intelligence task-solving abilities.

For engineering, research, and data science teams, this type of model can become a working tool for proof-of-concept in optimization, analysis, or structural reasoning, provided that careful evaluation of costs (premium subscription, cost per query) versus concrete benefits in existing pipelines is conducted.

—

Anthropic uses AI as an interviewer for users

Anthropic is launching a pilot program in which it uses an AI agent to interview users about their experience with AI, in 10-15 minute conversations. The questions target both ideal support needs and concerns about how AI could be developed or used contrary to personal values, as part of social research on the impact of models on people.

The experiment is relevant for how companies can use AI not just as a product but also as a UX research tool at scale. At the same time, it raises ethical design questions: how consent is collected, how anonymity is maintained, and how bias is prevented when a model asks users about itself.

—

Yahoo Sports uses AI for real-time NFL game recaps

Yahoo Sports has launched, in beta, the "Game Breakdowns" feature in its app, which automatically generates summaries, timelines of key plays, and question suggestions for NFL games. The system combines statistical data, user reactions, and Yahoo editorial content, using AI models to deliver a real-time overview, initially available for Yahoo Fantasy Plus subscribers.

For media and sports product developers, this is an example of how AI can become a layer over raw data, transforming box scores into an interactive experience. The direction is also relevant for other verticals with "real-time" data – stock markets, IoT, operational monitoring – where an AI agent can extract the "story" from event streams and metrics.

—

Cloudflare blocks AI bots and raises questions about the internet economy

Cloudflare has begun to block or limit access to certain crawlers associated with AI models, arguing that the traffic generated by them consumes significant resources without providing direct value to website owners. The article discusses the possibility of a "permission economy" emerging where access for AI training will be conditioned on explicit agreements, licenses, or payments.

For companies with relevant online intellectual property, Cloudflare's decision draws a line: public data is not necessarily implicitly available for AI training. From a technical perspective, new needs arise: granular bot management, dedicated traffic analysis for AI crawlers, and potential integration with platforms that monetize access to content APIs.

—

A "Fitbit for the brain" against chronic stress has launched

A startup is developing a wearable device based on EEG (electroencephalogram) that monitors brain activity and attempts to identify and reduce chronic stress. The idea is based on using high-frequency beta waves as a marker for psychological stress, and the product combines monitoring hardware with an AI software layer for data interpretation and personalized recommendations.

For digital health professionals and those building wellbeing applications, this example shows how century-old technologies (EEG) can be repackaged into consumer experiences, with AI as an intermediary between the raw signal and usable insight. At the same time, there is a need for clear security and confidentiality standards for extremely sensitive data.

—

Apple and Google warn users in over 150 countries about state-sponsored cyberattacks

Apple and Google have sent new rounds of notifications to users globally, warning them that they may be targets of state-sponsored cyberattacks. Apple states that it has issued alerts in over 150 countries since the start of this program, without detailing which groups are involved or how many users were targeted in the recent campaign.

For organizations, these notifications are a reminder that the risk of spear-phishing and device compromise is not just theoretical, especially for high-profile individuals (leadership, researchers, journalists, activists). Beyond software updates, implementing strict device management policies, access segmentation, and regular security awareness training becomes critical.

—

Future of Life Index: AI companies' safety practices lag behind emerging standards

A new report from the Future of Life Institute evaluates the safety practices of major AI companies – including OpenAI, Anthropic, xAI, and Meta – and concludes that they are "far below" emerging global standards. The assessment, conducted by a panel of independent experts, suggests a lack of robust strategies for controlling superintelligent systems and managing systemic risks.

For governments, regulators, and enterprise clients, the report is a signal that reliance on self-regulation cannot be solely counted on. At the organizational level, AI adoption should also be accompanied by clear contractual requirements regarding safety (testing, auditing, transparency) and internal risk management mechanisms that do not rely solely on supplier statements.

—

Google: AI search is an "expansion moment" for the web

A Google executive responsible for Search stated in an interview that integrating AI into results (including direct answers) is an "expansion moment" for the internet, not a threat to publishers. He claims that "outbound clicks" from Google to sites remain largely stable and that new ways of searching – through camera, complex queries, conversations – can increase "the size of the pie".

For software, content, and e-commerce professionals, the message is ambivalent: on one hand, Google asserts that traffic is not dramatically decreasing; on the other hand, the changing way users consume information (through AI summaries) requires adapting content strategies, structuring data to be easily retrievable, and maintaining brand visibility even when the primary response is generated by a model.

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