IT News Review by Control F5 Software: A single prompt can steer ChatGPT towards authoritarian ideas

A single prompt can steer ChatGPT towards authoritarian ideas

A study cited by NBC News shows that ChatGPT can be prompted to generate authoritarian-leaning responses through a single initial prompt. Researchers analyzed how the initial wording influences the direction of the conversation, even when subsequent questions are neutral or general.

The results indicate that the initial prompt functions as a persistent interpretive framework. Once a certain perspective is set, the model tends to maintain the same logic and produce responses aligned with that angle, without the need for repeated inquiries from the user.

For companies and organizations integrating LLMs into products or internal processes, the study emphasizes the importance of systematic testing of prompts and control mechanisms. Without such measures, model behavior can deviate in ways that are difficult to anticipate.

AI agents are not yet ready for office work

A new benchmark tests AI agents on real tasks from the professional environment, including consulting, investment banking, and the legal field. The goal is to evaluate these systems' ability to manage workflows, not just isolated tasks.

Tests show that agents encounter major difficulties in managing context, prioritizing tasks, and completing ongoing activities. In many cases, models get stuck or produce incomplete results when faced with processes that require continuity and adaptation.

For organizations, the conclusion is pragmatic: AI agents can be useful as support tools, but they cannot yet replace office work. Their adoption requires realistic expectations and rigorous evaluations before integration into production.

Google offers free SAT simulations generated with Gemini

Google has launched practice tests for the SAT exam, generated with the help of the Gemini model and available for free. Users can take full simulations, receive estimated scores, and explanations for incorrect answers, directly through the AI interface.

The company states that it has collaborated with education partners to ensure that the structure and difficulty level are close to those of the real exam. The goal is to expand access to preparation resources for students who do not have access to traditional tutoring.

This initiative highlights the increasingly active role of AI in education. For the software industry, the example shows how generative models can be integrated into products with a direct impact on learning and assessment processes.

Coordination is the next frontier for AI, says Humans

The startup Humans is developing an AI model focused on coordination among multiple users, not just individual conversations. The company starts from the idea that many organizational problems arise from a lack of alignment, not from a lack of information.

The founders come from organizations like OpenAI, Anthropic, and DeepMind and argue that current models are limited in contexts involving collective decisions, competing priorities, and different interaction histories.

For the industry, this approach suggests an evolution of AI towards systems that influence collaboration and workflows. Implementing such models raises challenges related to governance, accountability, and decision control.

Sci-fi writers and Comic-Con distance themselves from generative AI

Several organizations in the science fiction area have announced clear restrictions regarding the use of generative AI. The Science Fiction and Fantasy Writers Association has decided that works written entirely with AI are not eligible for awards.

Authors can only use AI as an auxiliary tool, with the obligation to declare this. The decisions reflect concerns related to copyright, originality, and the impact of AI on creative work.

For platforms developing creative tools, the message is clear: transparency regarding AI usage becomes an explicit requirement. Products must provide clarity on the role of AI in the creative process, not just technical capabilities.

ChatGPT begins to cite Grokipedia, the encyclopedia associated with xAI

TechCrunch reports that ChatGPT has started to include information from Grokipedia, an AI-generated encyclopedia associated with the company xAI. The publication mentions that Grokipedia has previously been criticized for errors and controversial claims.

The case brings back into discussion how language models select and prioritize information sources. OpenAI states that it uses a wide variety of public sources, without detailing the exact selection criteria.

For users and product teams, the situation underscores the importance of verifying sources and quality control mechanisms. In critical applications, reliance on unvalidated sources can have real consequences.

Apple may present a Siri based on Gemini

A report cited by TechCrunch suggests that Apple is set to unveil an updated version of Siri, built on the Gemini model, at an event scheduled for early next year. The new Siri would be closer to an AI chatbot, with extended support for text and more complex conversations.

Information indicates a shift in strategy from Apple's previous positions, which emphasized the discreet integration of AI into the system, not explicit conversational interfaces. Competitive pressure from other AI providers seems to have accelerated this decision.

For the Apple ecosystem, a Siri based on generative models raises important questions about privacy, local versus cloud processing, and control over access to personal data. These aspects are critical for enterprise and consumer adoption.

Gmail has had issues with spam and message classification

Google has confirmed the existence of an incident in which Gmail incorrectly classified legitimate emails as spam and allowed unwanted messages to reach inboxes. The problem affected users in multiple regions and was reported on the Google Workspace dashboard.

Users reported disruptions in communication flows, including moving messages between folders and delays in delivery. Google later stated that the issue has been resolved, without providing additional technical details.

For organizations, the incident highlights the strong dependence on automated filters and the importance of monitoring critical email channels. In business contexts, such errors can have a direct impact on operations and customer relationships.

Google Photos introduces AI-generated meme creation

Google Photos has launched an experimental feature that allows users to generate memes using their own photos. The AI combines personal images with predefined templates, resulting in easily shareable content.

The company specifies that the feature is aimed at recreational use and that results may vary depending on the quality of the photos. The launch is initially limited to certain markets and selected users.

This functionality illustrates how AI is becoming a standard component in consumer applications. For the software industry, the example shows how generative capabilities are directly integrated into existing products without radically changing the core experience.

DeepMind CEO criticizes OpenAI's rush to introduce ads

Demis Hassabis, CEO of DeepMind, stated that he is surprised by OpenAI's intention to introduce ads in ChatGPT. The comment was made in the context of discussions about the financial sustainability of large-scale AI services.

Hassabis suggested that monetization should be approached carefully, so as not to compromise product quality or user trust. The statement reflects strategic differences among major players in the AI field.

For the industry, the topic of ads in conversational interfaces raises issues related to transparency and clear separation between sponsored content and generated responses. These aspects will influence long-term adoption.

Anthropic updates principles for Claude

Anthropic has published a revised version of the document defining the ethical and safety principles of the Claude model. The document explains how the AI is guided by written rules, not just human feedback.

The approach, known as "Constitutional AI," aims to provide more predictability and transparency in model behavior. The update clarifies the limits and priorities in user interactions.

For companies, such frameworks become relevant in risk assessment and compliance. Documents of this type can be used as references in audits and in decisions regarding the acquisition of AI solutions.

YouTube will allow creators to use AI-generated images in Shorts

YouTube has announced that it will allow creators to generate Shorts using an AI version of their own image. The functionality is accompanied by tools that provide control over how the likeness is used.

The platform is also developing detection systems for unauthorized use of creators' faces or voices. The stated goal is to protect identity and reduce abuses related to deepfakes.

For the content ecosystem, this initiative indicates a formalization of AI usage in video production. Integrating AI with rights and consent mechanisms becomes essential for large platforms.

Users will spend more on apps than on games by 2025

A report cited by TechCrunch shows that by 2025, global consumer spending on non-game apps will surpass mobile game spending for the first time. The growth is mainly attributed to the accelerated adoption of AI-based applications.

Data indicates a significant increase in revenue generated by generative AI applications, both through subscriptions and in-app purchases. The use of these applications has increased not only in download volume but also in time spent in the app.

For the industry, this change marks a maturation of the mobile market. AI is no longer just an experimental feature but a product category that generates consistent revenue and influences monetization strategies.

Samsung begins production of HBM4 memory for AI

Samsung has announced the start of HBM4 memory production, amid increased demand from the AI industry. The company warns that the chip shortage may continue in the coming years as data centers expand.

HBM4 is used in AI accelerators and high-performance computing systems, being essential for training and running large models. Prioritizing this segment reflects the market's shift towards AI-dedicated infrastructure.

For companies building AI solutions, memory becomes a strategic factor. Infrastructure planning must take into account supply chain risks and the impact on medium-term costs.

Lenovo adopts a multi-LLM strategy for its products

Lenovo has announced that it will collaborate with multiple language model providers instead of developing its own LLM. The strategy aims for flexibility based on region, costs, and compliance requirements.

The company intends to integrate AI into devices and services, using different models depending on the context of use. This approach reduces dependence on a single provider and allows for rapid adaptation to market changes.

For software architecture, this strategy highlights the importance of an orchestration layer. Managing multiple models requires clear security policies, routing, and data control.

Autodesk reduces staff to invest in AI and cloud

Autodesk has announced layoffs affecting approximately 7% of its workforce as part of a strategic reorientation towards investments in AI and cloud. The company aims to allocate more resources for the development of future products.

The decision fits into a broader trend in the software industry, where mature companies are optimizing costs to fund initiatives deemed critical for competitiveness.

For clients and partners, such changes can influence roadmaps and delivery pace. Monitoring strategic direction becomes important for assessing long-term impact.

OpenAI launches a global initiative to increase AI adoption

OpenAI has announced a global initiative aimed at increasing the use of AI in everyday life, including in education, health, and emergency preparedness. The company collaborates with governments and public institutions.

The program includes adapting products to local contexts and investing in infrastructure, based on the needs of each region. The stated goal is to reduce technology access gaps.

For the software industry, the initiative indicates an increasingly close integration between AI and public policies. Compliance and data sovereignty requirements will become increasingly relevant.

OpenAI introduces age prediction in ChatGPT

OpenAI has launched an age prediction system in ChatGPT, aimed at the automatic application of protections for minor users. The system uses behavioral signals and account-level information.

Users incorrectly identified can contest the classification through verification processes. OpenAI states that the measure is necessary to improve safety and compliance with regulations.

For the industry, this approach shows that minor protection is no longer based solely on declared age. AI solutions will require clear transparency and contestation mechanisms.

Android 17 may introduce blur effects at the interface level

Information emerging in the Android community suggests that Android 17 will introduce blur effects at the interface level as part of a broader visual update. The change would target system elements and transitions between screens.

Although details are limited, such effects imply additional graphic resource consumption and may influence performance on older or less powerful devices. At the same time, blur is used to enhance visual hierarchy and readability.

For developers and product teams, changes at the UI level in the operating system require careful testing. Compatibility, accessibility, and user experience become critical factors in adopting new Android versions.

Gmail recommends checking settings after recent issues

A Yahoo article draws attention to some Gmail settings that may influence how messages are classified, in the context of recent incidents related to spam and misclassification. Users are encouraged to review their configurations.

The material highlights the importance of custom filters and security options, especially for accounts used for professional purposes. Adjustments can reduce the risk of important messages ending up in spam or being delayed.

For organizations, the article serves as a useful reminder that email delivery does not depend solely on the provider's infrastructure. Internal settings and email policies remain an essential component of operational hygiene.

Facebook is affected by a wave of low-quality AI content

The accelerated growth of AI-generated content on Facebook is a phenomenon that affects feed quality and user experience. The large volume makes it difficult to differentiate authentic content.

The low cost of content generation has led to an explosion of repetitive or low-value informational posts. Platforms are facing increasing pressure on moderation and recommendation systems.

For the software industry, the case highlights a structural problem: distribution and control are as important as generation. Without effective filtering mechanisms, AI can quickly degrade the digital experience.

Ring introduces video clip integrity verification

Ring has launched Ring Verify, a feature that allows users to verify whether a video clip from a security camera has been altered after recording. The system confirms the integrity of the file, not its content.

If a clip fails verification, users are informed that the file has been altered, without details about the nature of the modification. The feature is intended to increase trust in shared video materials.

For the security field, the difference between integrity and authenticity is essential. Ring addresses the first problem, but detecting deepfakes remains a separate challenge for the industry.

Internal documents show Google's strategy for Chromebooks in education

Internal Google documents suggest that education is seen as a strategic channel for long-term loyalty. Chromebooks are presented as an entry point into the Google ecosystem.

The strategy targets both hardware and associated software services, such as Google Workspace. Early adoption is correlated with long-term user loyalty.

For the industry, the topic raises questions related to ethics, data protection, and the use of AI in the educational environment. Regulatory pressure in this area is expected to increase.

Microsoft integrates AI features into Paint and Notepad

Microsoft is testing new AI features in classic applications like Paint and Notepad, available to users in the Windows Insider programs. The updates include content generation and editing assistance.

The updates show how AI is becoming part of the core tools of the operating system, not just specialized applications. The focus is on fast and accessible functionalities.

For developers, this direction changes user expectations. AI becomes an implicit standard, and product differentiation shifts towards integration and experience control.

American lawmakers call for transparency regarding data used to train AI

A legislative proposal suggests that authors and creators should be able to find out if their works have been used to train AI models. The initiative, known as the TRAIN Act, is bipartisan and aims to increase transparency in the industry.

The proposal comes in a context where disputes over copyright and datasets are becoming more frequent. Creators claim a lack of visibility into how their content is collected and used by AI companies.

For companies and enterprise clients, such a law could change the criteria for selecting AI providers. Data provenance, documentation, and auditability would become essential factors in the acquisition and implementation of AI solutions.

Substack tests a TV app for newsletters

Substack has launched a beta version of a TV app, expanding the distribution of newsletters to large screens. The initiative marks an evolution of the platform beyond the classic email and web format.

The app is designed for video and audio content, reflecting creators' growing interest in multimedia formats and more interactive experiences. Substack thus expands its positioning from publishing to complex digital media.

For publishers and media companies, the move suggests a change in infrastructure and strategy. Multi-platform distribution is becoming increasingly important, and integrating text with video and audio requires new production and monetization models.

Signs that the AI market is entering a maturation phase

An article from TechNewsWorld analyzes recent statements from the AI industry and suggests that excessive enthusiasm is beginning to temper. The focus is gradually shifting from spectacular promises to concrete and sustainable results.

The material highlights the increasing pressure on companies to demonstrate real value, operational efficiency, and cost control. Investors and clients are becoming more attentive to implementations, not just demos.

For organizations, this context favors pragmatic approaches. AI projects must align with clear business objectives, and success depends on proper integration into existing processes, not just technological capabilities.

Bret Taylor warns of a possible correction in the AI market

Bret Taylor, president of OpenAI, stated that the AI industry may go through a correction period after the wave of enthusiasm and accelerated investments. The comments were picked up by CNBC in the context of discussions about sustainability.

Taylor emphasized that not all AI initiatives will justify massive investments and that expectations regarding productivity may need to be adjusted. High infrastructure costs remain an important factor.

For the business environment, the message is one of caution. AI remains strategic, but implementations must be evaluated realistically, focusing on long-term value and cost control.

Meta suspends access for teenagers to AI Characters

Meta has announced the suspension of access for teenagers to AI Characters across all its platforms globally. The decision comes amid concerns about child safety and the use of AI in sensitive contexts.

The company is working on a "PG-13" version, which will include parental controls and clear content restrictions. AI Characters will be limited to topics considered appropriate for minors.

For the industry, the measure indicates an increase in regulatory pressure. AI products intended for the general public must integrate child safety directly into design and functionality.

AI models can develop emergent behaviors resembling personality traits

A study cited by Live Science shows that AI models can develop distinct behavior patterns when interacting freely, without predefined objectives or roles. Differences arise even among identical agents.

Researchers observed that repeated interactions and internal memory can lead to variations in responses that resemble personality traits. The phenomenon is described as emergent, not explicitly programmed.

For AI products, the conclusion is relevant for safety and consistency. Behavior is determined not only by the prompt but also by architecture, context, and ongoing interaction.

Apple is analyzing the transformation of Siri into a complete AI chatbot

Apple is working on transforming Siri into an AI chatbot, closer to the conversational experiences offered by competitors. The project is expected to be integrated into a future version of iOS and allow for both voice and text input.

The change marks a repositioning from Apple's previous strategy, which favored "invisible" AI integrated into the system. The pressure exerted by the rapid adoption of generative chatbots seems to have accelerated this decision.

For the Apple ecosystem, a conversational Siri raises important challenges related to privacy, data access, and permission control. These aspects are critical for maintaining user trust and for adoption in the enterprise environment.

OpenAI may launch its first hardware device in 2026

According to TechCrunch, OpenAI intends to launch its first hardware device in 2026, with indications that it could be a pair of earbuds. The project is associated with the design team led by former Apple design chief Jony Ive.

The device would emphasize natural interaction with AI and local processing of tasks, reducing dependence on the cloud. Details remain limited, and the company has not officially confirmed the final form of the product.

For the industry, such a move would change OpenAI's positioning from a software provider to a company that controls both software and hardware. Vertical integration may influence distribution methods and user experience.

Study: chatbots used for personal support are associated with anxiety

A study cited by NBC News indicates an association between the use of chatbots for personal support and higher levels of anxiety and depression. The research analyzes usage patterns without establishing a direct causal relationship.

Authors emphasize that users who frequently rely on chatbots for emotional support may reduce human interactions, which can exacerbate negative states. The study does not claim that AI is the direct cause of these issues.

For companies developing such products, the results highlight the importance of clear limits and responsible positioning. Chatbots should not be presented as substitutes for professional or human support.

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