📰 English IT Daily · 2026-06-19

A researcher says they found about 10,000 GitHub repositories that appear to distribute Trojan malware. The discovery began with a simple search. When the researcher checked whether search engines had indexed one of their own projects, they found another repository with exactly the same name and description. It also copied the full commit history, but a recent change had added a link to a ZIP archive in the README file.

Later, the researcher found another similar case while browsing GitHub tags. After watching these repositories, they noticed a repeated pattern. Every few hours, the previous commit was deleted and the same new commit was pushed again. The only visible change was in the README, where a link to a ZIP archive had been inserted. The repositories were not forks, and they came from different contributors with different names, which made the campaign harder to spot.

According to the report, the archive usually contained a small set of files, including a command script, an executable file, another file with a random name, and a DLL. If the archive link was submitted to VirusTotal, it reportedly showed no threats. However, when the ZIP file itself was scanned, security tools detected a Trojan. A Trojan is a type of malware that looks harmless but can secretly run dangerous actions on a computer.

To find more cases, the researcher created a general search pattern and wrote a script to look for repositories with the same behavior. The pattern included copied commits, repeated commit replacement, README-only updates, and links to ZIP archives in new non-fork repositories. The report also said that GitHub support was contacted, but action took time. The case shows how attackers may abuse trusted developer platforms and why engineers should inspect unusual repository activity carefully.

Researchers at MIT have built a new operating system kernel called Fractal to study how modern computer chips really work. A kernel is the core part of an operating system that controls hardware and software resources. The team said that common systems such as Linux or macOS were not designed for this kind of research. As a result, scientists often have to change existing kernels by hand, which can make experiments unstable and difficult to repeat.

Fractal was created from the ground up with a different goal: to treat the hardware itself as the main object of study. The researchers wanted a cleaner view inside a processor, especially when studying small behaviors that can affect security. These details matter because some attacks, such as Spectre and Meltdown, depend on hidden actions inside the chip. If researchers cannot clearly observe those actions, it is harder to test systems and understand risk.

One early use of Fractal was a close study of branch predictors in Apple’s M1 processor. A branch predictor is a part of the CPU that guesses what code may run next so the processor does not waste time waiting. This guessing improves speed, but it can also create security problems through speculative execution, where the chip temporarily performs work before it knows the final result is needed. According to MIT, Fractal helped reveal previously unknown behavior in the M1.

The researchers also reported the first evidence that a speculative attack class called Phantom affects Apple Silicon. More broadly, the project shows that better research tools can lead to better understanding of hardware security. Instead of adapting general-purpose operating systems for specialized experiments, the MIT team built a system designed for careful observation from the start. Their work may help future researchers study processors in a more reliable and reproducible way.

Project Valhalla is a long-running OpenJDK effort to make some Java objects work more like primitive types such as int. In June 2026, Oracle engineer Lois Foltan confirmed that JEP 401, called Value Classes and Objects, was integrated into the main OpenJDK repository and is targeting JDK 28. This is an important milestone after about a decade of work, but the feature is planned as a preview and will be disabled by default at first.

The main idea of Valhalla is often described as “codes like a class, works like an int.” In Java today, most data is stored as reference types. That means a variable usually holds a pointer to an object somewhere else in memory, not the object itself. This design is flexible, but it can reduce performance. Reading data may require pointer indirection, and each object also needs memory for metadata in its header.

These costs become more serious at scale. A large array of small objects can create many separate heap allocations, and later the garbage collector must clean them up. Because the objects may be spread across memory, access can be less efficient than with simple primitive values stored closely together. Valhalla tries to improve this by letting developers write normal classes with clear field names and methods while allowing the JVM to store and handle some values in a more compact way.

Still, the current step does not mean the whole project is finished. Community members noted that this is only the first part of Valhalla, and even supporters describe the situation as nuanced. The integration itself is large, which shows how deeply the change affects the platform. For Java developers, JDK 28 could be the first real chance to test these ideas in practice and see how value classes may change performance, memory use, and API design in future applications.

ClickHouse, an analytical database, marked ten years in open source in June 2026. An analytical database is software designed to process large amounts of data quickly for reporting and analysis. According to the company blog, ClickHouse has become a very popular open-source database and now has more than 2,000 contributors. The anniversary article says this growth came from building the project in public and making it easy for people to learn from the code.

The blog explains that open source can mean different things. At the lowest level, code is only published for people to read. At higher levels, development happens in a public repository, outside contributors are accepted, and the project has clear contribution guidelines. The strongest form also includes open roadmaps, code review, testing, continuous integration, release cycles, user support, and documentation. The article presents ClickHouse as a project that tries to follow this more complete and transparent model.

A major theme of the post is education. The founder says the source code is written so other engineers can study it. He describes the code as modular, well-documented, and explained with comments when ideas become complex. The project is also presented as a place to learn modern C++ development, including newer language features as well as practical engineering work such as build systems, testing, and review processes. In this way, the repository is not only a product but also a learning resource.

The article also highlights experimentation and contributor support. Developers can open a pull request to test a new idea, even if it is not certain to be merged. The same testing and review standards are applied to these experiments. The post mentions possible areas such as memory allocators, compression libraries, hash tables, data formats, and sorting algorithms. It also says contributors are publicly credited, including in changelogs and inside the database itself. The message is that open source is not only about sharing code, but also about building a welcoming engineering culture.

A recent report says AMD has quietly removed a memory encryption feature from some consumer Ryzen CPUs after newer AGESA firmware updates. AGESA is low-level startup code that helps the motherboard and processor work together. The change was not clearly announced to users, so many people may not know that a security feature they expected is no longer available.

Memory encryption protects data stored in system RAM by turning it into unreadable code. In simple terms, it can make it harder for an attacker with physical access to read sensitive information directly from memory. This kind of protection is especially useful for devices that may contain passwords, private files, or company data. Without it, some users could face a higher security risk in certain attack situations.

The report suggests that the feature disappeared after newer firmware versions were installed. Firmware is software built into hardware, and BIOS updates often include new firmware from chip makers. Because many users install updates to improve stability, compatibility, or performance, they may have lost this protection without realizing it. According to the report, AMD engineers did not give clear public answers when asked about the reason for the change.

This situation raises wider questions about transparency in hardware security. When companies add or remove security features, users and IT teams need clear communication so they can assess risk and decide how to respond. For business users, even a small undocumented change can affect device policy, compliance, and trust. The case is also a reminder that firmware updates do not only improve systems; they can sometimes change the security model as well.

As artificial intelligence becomes part of daily work, many people are asking a new question: can AI slowly weaken human skills? This concern is growing in medicine, computer science, and other fields where workers now use AI tools for difficult decisions. A recent survey of health-care workers in the United States found that many nurses and physicians worry about losing important abilities if they depend too much on AI systems.

Early research suggests that this problem, often called deskilling, may already be happening. One study followed experienced physicians in Poland who perform colonoscopies, a procedure that uses a flexible camera to examine the inside of the body. They used an AI system that analyzed images in real time and marked possible adenomas, which are pre-cancerous growths in the intestine. The tool was available on some days but not on others.

After the physicians became used to the AI system, their performance without it became worse. Before the tool was introduced, they found adenomas in a higher share of procedures. After several months of AI use, their detection rate dropped on days when the system was unavailable. The researchers said continuous exposure to AI assistance might make clinicians less focused or less responsible when they must make cognitive decisions alone.

Experts say these findings do not mean AI should be rejected. AI can still improve speed, accuracy, and support in many tasks. However, the results suggest that professionals need to think carefully about which skills they want to maintain and which tasks they are willing to outsource to machines. Researchers also say more studies are needed, but the early message is clear: if people rely on AI too heavily, some hard-earned expertise might atrophy over time.

Valve has released SteamOS Linux 3.8 as a stable version for the Steam Deck. A stable release means the software has moved beyond testing and is ready for general use. For users, this usually brings more confidence in everyday performance. For developers and power users, it also shows that the platform is maturing and becoming more dependable over time.

SteamOS is a Linux-based operating system, which means it is built on the Linux software platform instead of Windows. An operating system is the main software that manages hardware, apps, and system resources. On a device like the Steam Deck, the operating system plays a big role in battery life, game compatibility, controls, and overall user experience. Even small system changes can affect how smoothly games run.

Although the source context does not list detailed features, a stable update generally suggests a mix of bug fixes, performance improvements, and better system reliability. Bug fixes remove software problems. Performance improvements help the device work faster or more efficiently. Reliability means the system behaves in a consistent way without unexpected issues. These areas are especially important for handheld gaming devices, where users expect quick startup, responsive controls, and solid game sessions.

The release is also interesting beyond gaming. It shows how a Linux-based platform can be developed for consumer hardware in a practical way. This matters to programmers because operating system updates can influence driver support, testing, deployment, and application behavior. In a broader sense, SteamOS 3.8 reflects the growing importance of software maintenance: not only creating new features, but also improving stability, compatibility, and long-term platform support.

A new security feature called Enterprise-Managed Authorization is now stable for the Model Context Protocol, or MCP. MCP is a way for AI tools and other clients to connect to external servers, tools, and data sources. The new extension is designed for enterprise use, where companies need stronger control over who can access which systems. According to the announcement, organizations can now manage authorization centrally and let users reach connected MCP servers through a single login.

The update tries to solve a common problem with the usual OAuth model. In many cases, each employee must connect and authorize every server one by one. This creates friction during onboarding and can slow down adoption inside a company. It also makes security management harder, because access depends on what each user approved separately. Another risk is that work and personal accounts can become mixed if there is no clear requirement to use a corporate identity.

With Enterprise-Managed Authorization, the company’s identity provider becomes the main decision-maker for MCP server access. Administrators define policy in one place, and users sign in through single sign-on with their existing work identity. The system can then grant or deny access based on group membership, role, and conditional access rules. Behind the scenes, the client gets an identity assertion, which is a signed proof of who the user is, and exchanges it for an access token. Because of this flow, users do not need to go through a separate consent screen for each server.

The result is a zero-touch setup for end-users. If administrators enable a server for the organization, approved users can access it automatically when they log in. This also gives security teams a more centralized audit trail and makes it easier to avoid accidental mixing of personal and enterprise accounts. The extension is being adopted by companies including Anthropic, Microsoft, and Okta, as well as a growing number of MCP servers. The broader goal is to make enterprise access to tools and data simpler, safer, and easier to scale.

A recent blog post warns that online fraud is becoming more organized and more believable, especially in the job market. The writer describes a possible scam in which a recruiter contacts a skilled tech worker about a great job. The company name seems real, the pay is attractive, and the interview process feels normal. After a friendly screening call, the candidate is asked to sign a simple NDA, or non-disclosure agreement, on an online legal platform before continuing.

The danger begins when the person signs in to that platform with a familiar single sign-on method such as a major email account. In the example, the login page looks real, and even the two-factor authentication step appears normal. But the attackers may be capturing the victim’s login session at the same time. A login session is the temporary access a website keeps after you sign in. If criminals steal that session, they may continue using the account without needing the password again.

The blog post says the scam does not end after the fake interviews. In fact, the interviews and later rejection may be part of the theater, designed to reduce suspicion. During that time, attackers can monitor email, download cloud files, and log in to other services connected to the same identity. They may also hide warning messages by changing mail filters. With enough personal data, they can try identity theft, open credit cards, or attempt access to financial accounts.

The article’s larger point is that modern scams now combine social engineering with realistic digital tools. Social engineering means tricking people into trusting a false situation. AI can make these operations more convincing by helping criminals write better messages and imitate normal business communication. For professionals in tech, the lesson is not that every recruiter is fake, but that even experienced users can be targeted by careful, patient attacks. Strong verification habits and caution around login flows are becoming more important.

Leaked financial documents suggest that OpenAI is still losing billions of dollars each year, even though its revenue is growing very quickly. According to reports based on audited statements, the company’s revenue rose sharply from 2024 to 2025. By the end of 2025, monthly revenue was said to be close to $2 billion. This shows strong demand for AI products, but it does not mean the business is already healthy.

The biggest pressure comes from expenses. Research and development, or R&D, remained higher than total revenue in both years. In simple terms, R&D is the money spent to build and improve new AI models. These costs likely include training large models, which requires huge amounts of computing power. The documents also show that OpenAI paid a large amount of its R&D costs to Microsoft in 2025.

Another major cost is the cost of revenue, which means the money needed to deliver the service to users. For AI companies, this often includes inference costs, or the computing used when models answer user prompts in real time. As more people use chatbots and other AI tools, these serving costs can rise quickly. Sales and marketing expenses also increased, adding more pressure to the company’s overall finances.

Reports say OpenAI’s operating loss increased from 2024 to 2025, although it improved slightly as a percentage of revenue. Its net loss in 2025 was much larger, but part of that appears to be a one-time accounting charge linked to its move toward a for-profit structure. The bigger question is whether AI companies can turn rapid growth into profit. To do that, they may need to control training costs, reduce inference costs, and prove clear business value to enterprise customers.