IT News Review by Control F5 Software: GPT-5 now rivals humans in key professions

OpenAI says GPT-5 now rivals human experts in key professions

OpenAI has presented the results of a new test called GDPval, designed to assess how well artificial intelligence models perform in economically valuable tasks compared to human professionals. The study covered 44 occupations from new industries, including health, finance, journalism, and production. GPT-5 was considered equal to or even better than domain experts in over 40% of cases, while Anthropic's Claude Opus 4.1 achieved an even higher score of 49%.

Although the tests were limited to written reports and did not cover the full range of tasks in the real world, the results show rapid progress. Just 15 months earlier, GPT-4o scored only 13.7% in similar comparisons. OpenAI says these improvements could allow employees to transfer routine tasks to AI and focus on higher-value activities, although the company emphasizes that artificial intelligence is not yet ready to replace humans in the entirety of a profession.

ECB accelerates digital euro development through new innovation platform

The European Central Bank (ECB) has presented the findings of the first stage of its innovation platform for the digital euro, highlighting how the digital currency could function in practice. The first round of tests focused on everyday payments, from person-to-person transfers to online shopping. The results showed that a digital euro could integrate into existing banking and commercial systems, providing faster and more secure transactions.

Based on these results, the ECB announced a second stage of experimentation. This will explore more advanced functionalities, such as offline payments and cross-border transactions. The goal is for the digital euro to become a reliable, efficient, and widely accepted form of payment across Europe, while maintaining confidentiality and financial stability.

Spotify eases rules for AI music to increase transparency and reduce spam

Spotify has updated its AI policy to clarify how artificial intelligence is used in music production and to combat spam uploads. The streaming giant will adopt an industry standard called DDEX, which requires record labels and distributors to disclose whether AI was used for vocals, instruments, or post-production. This way, listeners will better understand the role of AI in a piece, without oversimplifying the process.

The company is also launching a new anti-spam filter to detect and downgrade fraudulent or low-quality music, while simultaneously banning unauthorized voice cloning and deepfakes. With the rapid increase of AI-generated music on streaming platforms, Spotify states that it supports authentic and creative uses of AI but wants to block malicious actors who abuse the system.

Microsoft integrates Anthropic AI into Copilot

Microsoft has integrated Anthropic's AI models into its Copilot assistant, expanding beyond its reliance on OpenAI. From now on, business users can choose between OpenAI models or Claude Opus 4.1 and Claude Sonnet 4 from Anthropic. This move reflects Microsoft's strategy to diversify its AI ecosystem and reduce dependence on a single partner.

The two Claude models have different roles: Opus 4.1 is geared towards complex reasoning, programming, and architectural design, while Sonnet 4 is more suitable for routine development, tasks on large datasets, and content generation. This flexibility aims to provide companies with more tailored AI support within the Microsoft suite.

OpenAI launches ChatGPT Pulse for personalized morning briefings

OpenAI has launched ChatGPT Pulse, a new feature that generates personalized overnight reports and delivers them to users in the morning. These daily briefs include news, personal schedules, and selected recommendations—from sports updates to family vacation plans. Pulse is designed to make ChatGPT feel more like a proactive assistant than a reactive chatbot.

Initially available for Pro subscribers at $200/month, Pulse integrates with applications like Gmail and Google Calendar to highlight important messages and events. OpenAI emphasizes that Pulse is not meant to encourage endless scrolling, like social media applications, but to provide a focused set of useful information each morning. Over time, the company hopes to expand Pulse to more users and add advanced features such as reservations or email drafting.

Beware of "workslop": the hidden cost of low-quality AI results at work

Researchers from BetterUp Labs and Stanford Social Media Lab have coined the term "workslop" to describe AI-generated content that looks good on the surface but lacks real value. This type of output often forces colleagues to redo or massively correct work, creating inefficiencies instead of productivity. A survey showed that 40% of American employees received low-quality AI-generated content in the last month.

The study warns that workslop contributes to the lack of return on investment in AI for many companies. To prevent the issue, leaders are encouraged to establish clear rules regarding AI usage and promote thoughtful, intentional applications instead of uncontrolled automation.

YouTube Music tests AI hosts for trivia and commentary

YouTube Music is testing AI hosts that provide trivia, stories, and commentary related to the songs you listen to. The feature aims to offer a more interactive and engaging experience, similar to Spotify's AI DJ, launched two years earlier.

The experiment is being conducted through YouTube Labs, a new hub for AI functionalities open to all users, although only a limited group in the U.S. can initially test it. This adds to previous tests, such as conversational radio and AI-powered search. YouTube states that the goal is to explore how AI can enrich music discovery without overwhelming users.

Anthropic triples international workforce amid demand for Claude AI

Anthropic has announced plans to triple its international workforce and expand its applied AI team fivefold this year, driven by global demand for its Claude models. Nearly 80% of Claude's usage comes from outside the U.S., with particularly high adoption in South Korea, Australia, and Singapore.

Backed by Google and Amazon and valued at $183 billion, Anthropic has grown its enterprise customer base from under 1,000 to over 300,000 in just two years. Annual recurring revenue has increased from $1 billion at the beginning of 2025 to over $5 billion in August. To support this growth, the company will open new offices in Dublin, London, Zurich, and its first Asian office in Tokyo.

MRI Software targets a $10 billion valuation through sale or IPO

Private equity firms TA Associates, Harvest Partners, and GI Partners are analyzing a possible sale or initial public offering (IPO) of the American company MRI Software, a provider of real estate management software. The transaction, expected in the next 12 months, could value the company at up to $10 billion including debt. Goldman Sachs is advising the process after MRI attracted interest from buyers in the real estate and technology sectors.

MRI generates nearly $1 billion in annual revenue and approximately $400 million EBITDA, with a steady growth rate of 10% and over half of its revenue coming from outside the U.S. A sale would represent a major exit for current shareholders, with GI Partners on track to achieve a ninefold return and TA Associates seven times its initial investment.

Hackers stole data of 8,000 children from London nurseries

Hackers stole personal information of over 8,000 children from nurseries in London operated by Kido International. The group, known as Radiant, published samples of the stolen data on the dark web—including names, photos, home addresses, and family contacts—to demonstrate the breach. They threatened to publish thousands more files with data of children and employees if their demands were not met.

British authorities, including the Metropolitan Police Cyber Crime Unit, are investigating the incident, but no arrests have been made. The hackers claim to be located in Russia, but there is no clear evidence. The attack highlights the growing danger of ransomware groups targeting vulnerable institutions, raising serious data security and child protection issues.

Cisco launches software to connect quantum computers via cloud

Cisco has announced new software that connects quantum computers from different manufacturers into a single quantum cloud, facilitating the development of practical applications. The tool analyzes computational problems, breaks them down into smaller parts, and distributes them across various types of quantum machines—regardless of the technology they use.

While Microsoft, Google, and IBM focus on building their own quantum computers, Cisco positions itself as a connector. Along with its new chips for quantum networks, the company aims to simplify access for developers, masking the complexity of differences between various quantum hardware approaches.

Meta launches "Vibes" – an AI-powered video feed for creativity

Meta has introduced a new feature called Vibes, a feed dedicated to short videos generated by AI. Users can create videos from scratch, remix existing clips, or enrich content with visual effects and music. These creations can be shared directly in the Vibes feed or reposted in Instagram and Facebook Stories or Reels.

The launch is part of Meta's strategy to accelerate AI development on its platforms. By providing more accessible and interactive tools for video creation, the company hopes to stimulate creativity and compete with TikTok and other short video platforms.

EU prepares charges against Meta for failure to combat illegal content

Meta risks a charge from the EU for failing to sufficiently control illegal content on Facebook and Instagram. According to Bloomberg, preliminary findings from the European Commission show that the platforms do not have adequate mechanisms for users to report harmful posts.

Under the Digital Services Act (DSA), tech giants must implement stricter measures against illegal and harmful content. If the charges are confirmed, Meta could face fines of up to 6% of its global annual revenue. The company is also under investigation for failing to combat misinformation and misleading ads ahead of the 2024 European elections.

Microsoft unifies AI app stores into a single marketplace for business

Microsoft has merged its separate marketplaces for AI tools—one aimed at developers on Azure and another for business-oriented AI applications and agencies—into a single hub called Microsoft Marketplace. The launch begins in the U.S. and will be expanded globally in the coming months.

The new marketplace is aimed at corporate buyers, offering simple integration with Microsoft products and billing systems. Unlike consumer app stores, Microsoft will not charge commissions on sales but will require a publishing fee. All applications must pass strict security and compliance checks to be safely used in the enterprise environment.

The viral call recording app Neon shuts down after massive data leak

Neon, a viral app that allowed users to record calls and be paid by selling the recordings to AI companies, has been abruptly shut down after a major security breach. The bug allowed anyone to access phone numbers, transcripts, and audio files of other users, raising serious privacy concerns.

Despite attracting 75,000 downloads in a single day and quickly climbing the Apple charts, the app's servers were not properly secured. The founder shut down Neon after being alerted by TechCrunch but did not inform users about the extent of the incident. It is unclear whether the app will return or if Apple and Google will intervene.

Threads tests a "tagging" tool for more personalized feeds

Instagram is developing a feature for Threads that allows users to "tag" an account to signal what type of content they want to see more or less of in their feed. The idea is to give people more direct control over how the algorithm works, similar to what Elon Musk recently announced for X through his AI assistant, Grok.

The tool, still in prototype stage, would allow users to adjust their experience on Threads in real-time without having to navigate through settings. This reflects Meta's broader efforts to provide more transparency and flexibility in how feeds are curated, as both users and regulatory authorities demand more control over social media algorithms.

Google launches Mixboard, an AI app for mood boards

Google has launched Mixboard, a new app that allows users to generate and edit mood boards using artificial intelligence. Unlike Pinterest, where users must search for and manually add images, Mixboard allows starting from text prompts, which AI transforms into thematic boards. Users can edit, combine, or regenerate visuals using Google's Nano Banana image editing model.

Available as a public beta in the U.S. through Google Labs, the app is intended for brainstorming projects, such as interior design, event planning, or creative collages. Mixboard also adds text generation based on images, opening new creative possibilities. The app aligns with Generation Z's interest in digital collages, where similar tools have gone viral on TikTok.

Microsoft Photos will auto-categorize receipts and documents

Microsoft is testing an AI update for the Photos app in Windows 11, which automatically sorts certain types of images into folders. The feature will categorize receipts, screenshots, identity documents, and handwritten notes without users needing to organize them manually.

The AI can recognize and sort images based on their content, even if the text is in a language different from English. For example, a photo of a Hungarian passport will be placed in the "identity documents" folder. Although currently limited, Microsoft may expand functionality for more personalized sorting, such as pet photos or vacation pictures.

Meta brings controversial "Pay or Consent" model for ads to the UK

Meta is implementing its "pay or consent" model for ads in the UK, forcing Facebook and Instagram users to accept personalized ads or pay for an ad-free subscription. Prices start at £2.99/month on the web but rise to £3.99 on iOS and Android due to app store fees. Each additional account costs extra, and users must apply the same choice across all their accounts.

The model was previously rejected by EU regulators, who criticized it for offering users a "false choice." The fees in the UK are significantly lower than in Europe, where subscriptions started at €9.99/month. The measure highlights Meta's ongoing struggle with regulators over how it finances its platforms while also respecting user privacy.

Nvidia makes open-source Audio2Face AI for realistic voice animations

Nvidia is offering developers its Audio2Face tool for free, which allows for the generation of realistic facial animations for 3D avatars directly from audio input. The tool maps the acoustic features of speech to synchronize lip movement and facial expressions, bringing digital characters closer to realism in games, applications, and livestreaming.

Developers have already used Audio2Face in titles like Chernobylite 2 and Alien: Rogue Incursion. Along with models and SDKs, Nvidia is also releasing a training framework, allowing users to adapt it for various cases. By going open-source, the company hopes to expand usage in gaming, virtual production, and the metaverse.

Microsoft tests microfluidic cooling to increase chip and data center efficiency

Microsoft is experimenting with microfluidic cooling, a technique that directs cooling liquid through microscopic channels etched directly onto chips, inspired by natural vein patterns. Tests show that it can remove heat up to three times more efficiently than current cold plate systems, reducing GPU temperatures by 65%.

If the method is scaled, it could make AI-focused data centers more energy-efficient and allow chips to operate at higher power without overheating. By managing demand peaks without additional servers, the technology promises both cost savings and sustainability benefits—although Microsoft warns that it is still in an early stage.

Google DeepMind's Gemini robots can now search the web to solve tasks

Google DeepMind has enhanced its AI for robotics with Gemini Robotics 1.5 and Gemini Robotics-ER 1.5, allowing robots to plan complex tasks and use web search for tailored solutions. Robots can now do things like pack a suitcase based on weather forecasts, separate laundry, or sort recyclables according to local rules.

The system combines perception from the real world with online information. Robots can also "learn" from each other, transferring skills between different types of hardware, including humanoid robots. Developers will be able to access these models through the Gemini API in Google AI Studio.

The UK anti-fraud AI tool recovers £500 million and will be licensed globally

A new AI fraud detection system has helped the UK government recover nearly £500 million in public funds over the past year, marking the largest sum ever recovered in a single year by anti-fraud teams. Over a third of the money comes from fraud related to the Covid-19 pandemic, with the rest obtained from illegal local tax claims and illegal subletting of social housing. The system correlates data between departments and blocks suspicious activities, such as companies with Bounce Back loans trying to dissolve before repayment.

The tool, called Fraud Risk Assessment Accelerator, was developed by the Cabinet Office to prevent a repeat of the massive fraud during the pandemic. It scans new government policies and procedures to detect vulnerabilities before they can be exploited. Ministers say the recovered funds will finance the hiring of teachers, nurses, and police officers.

The UK now plans to license this tool internationally, with the U.S., Australia, Canada, and New Zealand among the first countries to adopt it. However, the decision could spark criticism from civil rights groups, already unhappy with the use of AI in combating fraud. Reports from organizations like Amnesty International have warned that such systems can discriminate based on criteria such as age, disability, or nationality.

Google opens real data doors for AI with new MCP Server

Google is making it easier for AI systems to access real and reliable information. The company has launched the Data Commons Model Context Protocol (MCP) Server, which allows developers, data scientists, and AI agencies to access verified statistics through natural language— a move that could change how AI models are trained and optimized.

Databricks commits to spend $100 million with OpenAI, as two of the largest tech startups join forces

Two of the world's most valuable startups— Databricks and OpenAI— are strengthening their partnership in the AI race.

On Thursday, Databricks announced that it will spend $100 million over several years with OpenAI, integrating GPT-5 and other models directly into its data analytics platform. This move makes it much simpler for Databricks customers to apply OpenAI technology to their enterprise data, without complex configurations, legal hurdles, or security issues that previously slowed adoption.

"It's a game changer," said Ali Ghodsi, CEO of Databricks. "Now, any Databricks user can run OpenAI models with a single click in the interface—at prices comparable to those directly from OpenAI."

Meta expands Teen Accounts globally on Facebook and Messenger

Meta is expanding the Teen Accounts feature for Facebook and Messenger users worldwide, after it was initially available only for teens in the U.S., U.K., Australia, and Canada. The safety-focused accounts, first launched on Instagram last fall, come with built-in protections and parental controls.

The expansion comes amid increased pressure from lawmakers, who have accused major social platforms of not providing enough protection for young users.

Robotics pioneer warns: the "bubble" of humanoid robots will not last

Rodney Brooks, one of the world's most respected robotics experts, has a blunt message for investors pouring billions into humanoid robot startups: "Throw your money out the window."

Brooks, co-founder of iRobot and a professor at MIT for decades, is deeply skeptical of companies like Tesla and Figure, which believe that showing robots videos of people at work can teach them dexterity. In a new essay, he dismisses this approach as "pure fantasy."

Google: 90% of tech employees now use AI at work

Artificial intelligence has rapidly become an almost universal tool for those working in the tech industry, according to a new study conducted by Google.

The research, conducted by Google's DORA division and based on 5,000 responses from professionals worldwide, shows that 90% of tech employees use AI in their work— a 14% increase from last year. Many rely on AI for tasks such as writing or refining code.

The rapid adoption comes amid a broader debate about the impact of AI on jobs and the economy. Meanwhile, Anthropic CEO Dario Amodei warned that AI could lead to increased unemployment. Other industry leaders have downplayed these fears, but there is evidence that entry-level jobs in software engineering are becoming harder to obtain, especially in the context of the wave of layoffs in the tech industry.

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