OpenAI has launched GPT-Red, a new artificial intelligence model specifically designed to simulate targeted attacks on large language models. This development aims to identify weaknesses and strengthen the security of the company's own systems before they reach the market.
The developer highlighted the rapid evolution of its technologies, justifying the need for this new defensive strategy, stating that simulating cyberattacks is fundamental. The primary focus of GPT-Red lies in combating prompt injection, which consists of crafted texts designed to trick an AI into ignoring its safety guidelines.
With the increasing integration of assistants into local files, browsers, and third-party applications, a malicious instruction can be disguised in common communications, such as emails or web pages, executing in the background without the user noticing, constituting a type of rising invasion in the technology sector.
The model operates by simulating a human hacker. It initiates a command on the tested AI, analyzes the generated response, and autonomously refines subsequent messages. This process repeats until the tool achieves a predefined objective, such as forcing the leakage of confidential data to an external server.
To increase its proficiency, OpenAI employs reinforcement learning. In this methodology, the system learns through trial and error, receiving rewards for successful attacks and penalties for failures, until it discovers the most effective invasive tactic. Observations from discussions on Reddit indicate that this establishes a cycle of continuous evolution: as defensive AIs become more resistant, GPT-Red is compelled to create more sophisticated and creative invasion methods.
The data presented demonstrates that the automated approach significantly surpasses manual effort. In internal tests focused on indirect injection, GPT-Red succeeded in 84% of the proposed scenarios, while human experts managed to overcome blocks in only 13% of attempts.
To validate the tool's effectiveness, OpenAI applied GPT-Red to commercially used software. One target was the system of an AI-controlled vending machine. The OpenAI model fully met the test objectives: it managed to reduce the price of an expensive item to the minimum allowed value of US$0.50, buy a $100 item for this negligible price, and even cancel a legitimate customer's order. After this experiment, the flaws were reported and corrected by those responsible.
In another test, GPT-Red attacked a programming agent based on GPT-5.4 mini to assess its security against data extraction. The invasive tool managed to force the system to leak sensitive information with great ease. Furthermore, the AI identified a new attack tactic with a success rate exceeding 95% against the previous model, GPT-5.1. By integrating the data generated by GPT-Red into the training of recent products, OpenAI ensures that the newly released GPT-5.6 Sol exhibits flaws in only 0.05% of direct injection attempts.
Due to its high potential for manipulation, OpenAI confirmed that GPT-Red will not be made available to individual customers or developers. This restriction serves as a safeguard to prevent malicious actors from accessing the program's capabilities and using them to automate cyberattacks against other platforms. The company's plan is to keep this new artificial intelligence operating strictly behind the scenes to protect its own infrastructure.
Samsung is scheduled to hold the Galaxy Unpacked 2026 event on Wednesday, July 22nd, starting at 10 am. At this event, the company is expected to reveal several innovative products, with a special focus on the Galaxy Z Fold 8, which will have a format described as a 'passport'.
Rumors suggest that the Galaxy Z Fold 8 will have closed dimensions of 129 x 82.2 x 9.8 mm and an expanded screen of 161.4 mm in width. This model would be the first from Samsung with a 4:3 aspect ratio, resembling a passport booklet. Its specifications include two rear cameras up to 50 MP, a 4,800 mAh battery, and support for 45-watt charging.
The line is also expected to introduce the Galaxy Z Flip 8, which is anticipated to bring few changes compared to the previous model. Additionally, there will be the introduction of two new Galaxy Watch devices and a smart glasses equipped with artificial intelligence.
There are speculations about a change in the line's nomenclature, where the new foldable would be named Galaxy Fold 8, and the successor to the Z Fold 7 would be called Z Fold 8 Ultra. Although there is this name change, rumors suggest that the Ultra device will not have a significant advancement in terms of functionality.
The Galaxy Z Fold 8 Ultra, which would be the most robust model, would maintain the three cameras up to 200 MP of the base model launched in 2025, using the same 5th generation Snapdragon 8 Elite chipset found in the Galaxy Fold 8 'passport'.
Regarding the Galaxy Z Flip 8, it should maintain the two cameras and the maximum battery capacity of 4,174 mAh from its predecessor. Confirmed improvements for this model include a reinforced hinge and the adoption of a Qualcomm SoC, similar to the one used in the Fold line.
Samsung is also expected to present the Galaxy Watch 9 and the Galaxy Watch Ultra. Both are expected to receive improvements in battery life and incorporate the AI insights feature from the Samsung Health app, presented in June. While the Watch 9 should maintain standard features, the new Ultra 2 promises the most powerful battery in the line, with 800 mAh, focused on outdoor activities.
Concerning the smart glasses, a prototype was previously seen, allowing interaction with Gemini, as tested by journalist Scott Stein from CNET. The design of these glasses would be under the responsibility of Gentle Monster.
The Galaxy Unpacked 2026 event will be broadcast live on the Samsung website and YouTube on Wednesday, July 22nd, starting at 10 am (Brasília time). Tecnoblog will cover the event in person, directly from London, England, by invitation from the company.
Google announced new features for the AI Mode in Search this Thursday (16th), integrating it with Canva, YouTube Music, and Instacart. These additions expand the use of artificial intelligence to facilitate users' daily tasks.
The implementation of these improvements began in the United States during July. The corporation's intention is to make AI features closer to usual activities, allowing users to trigger tools from other applications directly through the search interface, eliminating the need to open each service individually.
With these new connections, AI Mode now assists in creating playlists, developing visual content, and organizing purchases, all based on commands provided in natural language within Google Search itself.
The partnership with YouTube Music allows users to request customized playlists based on criteria such as musical style or desired mood. After processing the request, the system suggests a playlist that can be viewed in the music platform's application.
Regarding Canva, the AI tool offers support in creating visual templates, such as materials for events. Google clarified that this technology has the capacity to use the user's calendar data, when available, to adapt the templates to the specific context of the event.
Additionally, the link with Instacart allows orders placed via AI Mode to be converted into shopping lists within the application. This functionality can also take into account scheduled commitments in the calendar, such as meetings or events, to suggest items relevant to the occasion.
This expansion is part of Google's Personal Intelligence initiative, a feature that enables Gemini to use information from various applications to enrich its responses. The company began increasing this capability throughout 2026.
In addition to these integrations, AI Mode received more sophisticated search features. Among them is a smart search box designed to adapt to more elaborate queries and accept multiple input formats, including videos, images, files, and Chrome browser tabs.
The company also expressed its intention to increase the number of partners connected to the system in the future, maintaining a strategic focus on converting artificial intelligence into a cohesive tool across various spheres of digital routine.
Researchers from the University of Sheffield (England) have developed a new theory suggesting that the mysterious substance holding galaxies together may be in harmony with a hidden fifth dimension. The goal of this research is to help explain one of the greatest mysteries of modern science—the true nature of dark matter.
Dark matter has been studied by scientists for decades and inspires works of science fiction. For example, it appears as vortices capable of destroying planets in the film 'Star Trek,' or as 'dust' sustaining the multiverse in Philip Pullman's fantasy trilogy 'His Dark Materials.' Despite this, it remains one of the most unresolved problems in physics.
Although this substance has never been directly observed, scientists are confident in its existence due to its intense gravitational influence. It acts as an invisible 'cosmic glue,' ensuring the cohesion of galaxies, but its composition and properties remain unknown.
The hypothesis that this substance might exist in an additional and hidden dimension has been discussed in recent years. However, researchers from the University of Sheffield claim to have made a step forward with a new theoretical model. This study, published in the journal Physical Review D, offers an explanation for the behavior of dark matter and the difficulties in detecting it.
According to this model, dark matter exists in an extra dimension hidden near a force-carrying particle known as the dark photon. The theory posits that the specific geometry and shape of this dimension lead to the precise alignment of these particles' masses. This alignment generates a phenomenon called dark matter resonance, which researchers compare to the operation of a musical instrument vibrating strongly when reaching the correct note.
Dr. Yu-Dai Tsai, a senior researcher at Dorothy Hodgkin's Royal Society at the University of Sheffield, told Phys.org that dark matter resonance is already considered an important idea for understanding both its origin and detection methods. He noted: 'Dark matter resonance is already known as a powerful idea with the potential to change our understanding of how dark matter was produced in the early Universe and how we search for it today.'
According to the scientist, previous models treated this resonance merely as a pre-assumed hypothesis. Tsai added: 'But many previous models of resonant dark matter viewed the resonance as an assumption. This work provides a possible deeper origin for it: the resonance may stem directly from the geometry of hidden dimensions.'
Tsai also stated that this mechanism could explain why dark matter interacted much more intensely during fundamental periods of cosmic history, remaining practically undetectable today. He emphasized: 'This resonance can make dark matter interactions much stronger during critical epochs of cosmic history, such as in the early Universe. It is important that the model allows for these strong past interactions while explaining why dark matter seems so inert and difficult to detect today.'
The researchers also point out that although resonant dark matter and extra dimensions were studied separately in previous works, these models required extremely precise tuning of particle masses to function. The new study suggests that this alignment is not a coincidence nor dependent on manual adjustment, but arises naturally from the mathematical structure of the hidden dimension itself.
For Tsai, understanding this mysterious substance would mark a significant breakthrough for science. He stated: 'Understanding dark matter would be a profound breakthrough in humanity's knowledge of the cosmos and what it is made of.'
He added that the new model opens up new avenues for future research. 'Our study offers physicists new clear goals in the search for dark matter, while simultaneously linking two of the greatest ideas in fundamental physics: the mystery of dark matter and the existence of hidden dimensions.'
Beyond the implications for understanding the Universe, the researchers note that the search for dark matter stimulates the development of technologies with practical applications. These include ultra-sensitive detectors, cryogenic systems, low-noise electronics, and quantum measurement technologies, which can contribute to progress in fields such as medicine, computing, and global communications.
