📰 English IT Daily · 2026-06-22

The Linux kernel has removed the strncpy API after a long cleanup effort that lasted about six years. The work involved around 360 patches, showing how hard it can be to change old code in a very large software project. The kernel is the core part of Linux, and even small programming changes can affect many parts of the system.

The strncpy function is a C programming API used to copy text into a fixed-size buffer, which is a memory area with a limited length. It was designed to help prevent overflow, a situation where data writes past the end of memory. However, developers have found that strncpy can still cause confusion and bugs because its behavior is not always simple or predictable. In some cases, it may not add a final zero byte, which can create problems when software reads the string later.

Because of these risks, Linux developers have spent years replacing strncpy with safer and clearer alternatives. This kind of cleanup is important in security-sensitive code such as an operating system kernel. The goal is not only to reduce bugs, but also to make the code easier to review and maintain. When developers use APIs with clearer behavior, it becomes easier to understand what the code is trying to do.

This change also shows how modern software engineering often means improving old code instead of only adding new features. Large open-source projects must balance compatibility, performance, and safety. Removing a widely used API takes planning, testing, and patience. For programmers, the Linux effort is a useful example of technical debt reduction: small risky patterns can remain in code for years, and fixing them may require steady work across many releases.

Linux developers often need a good way to handle asynchronous I/O, which means working with many network or file operations without blocking the whole program. For years, epoll was the standard tool for this job. It became popular because it helped servers manage many connections at once. A recent article compared epoll with io_uring, a newer Linux feature, and explained why some developers are moving from one system to the other.

The main idea of epoll is simple: it tells a program when I/O is possible. For example, it can notify a server that a socket is ready to read from or write to. After that, the program still has to call read() or write() itself. This design works well, but it depends heavily on syscalls, which are requests from a user program to the Linux kernel. When a server handles many connections, these repeated syscalls and context switches can create serious overhead.

io_uring takes a different approach. Instead of saying that I/O is possible, it reports when I/O is done. It uses shared memory between the application and the kernel, organized in ring buffers. The program places requests into a submission queue, and the kernel returns results in a completion queue. In many cases, one system call can submit a batch of operations and receive a batch of completions. This can reduce overhead compared with doing separate calls for each event.

The article also notes that io_uring is not magic. By default, programs still need to call io_uring_enter() so the kernel can check the queue. There is also an option called SQPOLL, where a kernel thread watches the queue and can reduce syscalls even more, but it may use extra CPU. In short, epoll is older and simpler, while io_uring offers a more modern model that may improve performance. For engineers building high-performance servers, understanding this trade-off is becoming increasingly important.

A Danish privacy activist, Lars Andersen, said police raided his home, according to a post shared online. Because the public source is limited, many details about the case remain unclear. However, the report has already drawn attention from people interested in digital rights, security, and freedom of expression. When police action involves a privacy advocate, it often creates public concern about whether the response was necessary and proportionate.

Privacy activists usually work on issues such as surveillance, personal data, and the power of governments or companies to collect information. In simple terms, surveillance means watching or tracking people’s activities, often through cameras, networks, or online services. Personal data includes information that can identify a person, such as names, addresses, or device records. Cases like this can quickly become sensitive because they touch both public safety and civil liberties.

For the technology community, the event is important because trust is a key part of security. If people believe that legal action is being used against activists or researchers without enough transparency, trust in institutions can weaken. Transparency means giving clear information about what happened and why. At the same time, authorities may say they must protect an investigation, which can limit what they can publicly explain in the early stage.

This case also highlights the wider debate about digital rights in modern society. Security is not only about stopping crime; it is also about protecting the rights of ordinary users, journalists, researchers, and activists. For IT professionals, especially those who manage systems and data, the lesson is to understand both technical security and the legal context around privacy. Even when facts are still developing, such incidents remind us that technology, law, and human rights are closely connected.

Minecraft: Java Edition 26.2 has been released with new gameplay content and an important graphics update. The version adds the Sulfur Cube mob, the Sulfur Caves biome, new sulfur and cinnabar block sets, and several new music tracks. It also adds a Friends List and more language support. For technology watchers, the most notable change is experimental support for rendering the game through Vulkan, along with a new “Graphics API” option in the video settings.

A graphics API is a software interface that lets a game talk to the computer’s graphics hardware. In simple terms, it helps the game engine send drawing commands to the GPU. Vulkan is a modern graphics API designed to give developers lower-level control and better efficiency than older systems in some cases. The Minecraft announcement says this is experimental support, which means it is an early version that may still have bugs, limited compatibility, or performance issues on some devices.

This update is also interesting because it shows how a very popular Java-based game continues to evolve at the system level. Java Edition is known for gameplay flexibility and a large modding community, but graphics changes are especially important because they affect performance, driver support, and cross-platform development. By adding Vulkan 1.2 support, the game may open new possibilities for future optimization, although players should expect testing and gradual improvement before the feature becomes fully mature.

Beyond graphics, version 26.2 adds creative mechanics through the Sulfur Cube. The mob can absorb blocks, change its behavior, and become a physical object that players can push or use in different ways. This creates new options for building, movement, and mini-games. Overall, the release combines gameplay design with low-level rendering work, making it notable not only for players but also for software engineers who follow how large games update their architecture while keeping the user experience fresh.

Mitchell Hashimoto said his family is pledging another $400,000 to the Zig Software Foundation. With this new pledge, their total promised support for the foundation will reach $700,000, including an earlier donation in 2024. He said Zig continues to earn his respect both as a technical project and as a community. Zig is a programming language, and its compiler is the software that turns human-written code into machine instructions.

In his post, Hashimoto pointed to steady progress in Zig’s 2026 development update. He said the project is working on difficult problems involved in building a high-quality language and compiler. He also praised Zig’s approach to maintainership, which means how a project is managed and cared for over time. He mentioned community efforts such as Contributor Poker and the project’s AI ban as examples of a clear philosophy about quality and participation.

The donation also comes during public debate about Zig’s strict no-LLM contribution policy. An LLM, or large language model, is a type of AI system that can generate text or code. Hashimoto said he uses AI heavily and has written about building real features with AI assistance. However, he also said he is frustrated by some negative effects of AI on open source. Even though his own views do not fully match Zig’s policy, he said he still respects the people, rules, and goals behind the project.

Hashimoto argued that open source is valuable partly because projects can be different from one another. In his view, they should be free to set unusual boundaries, create their own culture, and pursue quality in their own way. He described Zig as ambitious, practical, independent, and unusually serious about quality. He also said Ghostty exists in large part because Zig made it possible for him to build the kind of software he wanted. He ended by saying he is proud to support Zig again and encouraged others to donate if they can.

A recent case study from Bayer and Thoughtworks shows how companies can build more reliable agentic AI systems for real work. The project, called PRINCE, is a cloud-hosted platform for preclinical drug research. It was created to help researchers search and use decades of safety study reports more easily. In the past, teams often depended on keyword search and strict Boolean logic, which could not handle complex questions well.

PRINCE later developed into an intelligent research assistant. It uses agentic retrieval-augmented generation, or RAG, which combines language models with document search so answers are based on trusted internal data. It also uses Text-to-SQL, a method that turns natural language questions into database queries. This allows researchers to ask questions in everyday English and receive context-rich answers, instead of manually checking many separate reports.

The article says reliability did not come from the model alone. Engineers focused on context engineering, meaning they carefully shaped and routed information between specialized agents. Different agents had different jobs, such as clarifying user intent, planning steps, validating data, and writing final answers. The team also built harness engineering around the models. This included orchestration, recovery, and observability so the system could stay controlled, recover from errors, and be monitored in production.

Trust was a central goal of the system. The PRINCE platform emphasized transparency, explainability, and human-in-the-loop review so users could better understand and check AI outputs. The case study suggests that production AI needs governance and compliance as well as strong technical design. It also shows an important lesson for many industries: if organizations want useful agentic AI, they must combine model capability with evaluation, monitoring, and careful system engineering.

As artificial intelligence becomes part of daily work, many experts are asking an uncomfortable question: can AI slowly weaken the skills people spent years building? This concern is growing in medicine, computer science, and other fields where professionals now use AI systems for support. Surveys of healthcare workers suggest that many nurses and doctors are worried about becoming too dependent on these tools and losing important abilities over time.

One study in Poland gives an early warning. It focused on physicians who perform colonoscopies, a procedure that uses a flexible camera tube to examine the inside of the body. The doctors used an AI system that analyzed images in real time and marked possible adenomas, which are growths that can become cancer. Before the tool was introduced, the physicians found adenomas at a higher rate. After they became used to AI support, their detection rate fell on days when the system was not available.

Researchers say this could be a sign of deskilling, which means a person becomes less capable because a tool takes over part of the job. Even highly trained professionals may lose focus or confidence when they stop making as many decisions by themselves. According to the study authors, continuous exposure to AI can make clinicians less motivated, less focused, and less responsible when they need to work without assistance.

Experts say more research is needed before strong conclusions are made, but the issue is serious enough to deserve attention now. The main question is not whether AI should be used, but how people can use it without losing core expertise. Some researchers argue that workers should think carefully about which skills they want to maintain and which tasks they are willing to outsource. As AI spreads across industries, preserving human judgment may become as important as improving efficiency.

Anthropic has published support information about identity verification on Claude. Identity verification is a process that checks whether a user is a real person by confirming personal information. The company says this step may be required in some situations. It is part of a broader effort to improve account safety and manage access to AI services more carefully.

For many online platforms, identity verification is used to reduce fraud, abuse, and policy violations. In the AI industry, these concerns can include fake accounts, misuse of tools, and attempts to avoid platform rules. By verifying identity, a company can connect an account to a real user more reliably. This can support trust and make it easier to investigate suspicious activity.

At the same time, identity verification can raise privacy questions. Users may worry about how their personal information is collected, stored, and protected. Support pages about verification usually explain what is needed, when the process applies, and what a user can do if verification fails. Clear communication is important because people want to understand why extra checks are necessary and how the process affects their account.

The topic is important beyond one AI service. As AI tools become more widely used, companies are under pressure to balance easy access with stronger security controls. Identity verification is one possible control, but it needs careful design to avoid creating too much friction for legitimate users. For businesses and engineers, this development shows how product decisions, compliance needs, and user experience are becoming closely connected in modern AI systems.

A new article by Brandur Leach looks at a basic question for software founders in the age of AI: what is the smallest useful product that people will still pay for? He calls this the minimum viable unit of saleable software. The idea is important because large language models, or LLMs, have made it much easier to build internal tools. Tasks that once needed a team and a long schedule can now be done faster with AI support.

The article starts with a simple example. A company was paying a monthly fee for Jira, a popular task tracker, and decided to replace it with an internal tool built with help from Claude, an AI model. This shows how the old buy-versus-build decision is changing. In the past, building software outside a company’s core domain was often too expensive. Now, AI can lower the cost and reduce the time needed for an initial version.

However, the author argues that cheap does not mean free. LLM-built software still needs a human feedback loop. In other words, a person must review the results, ask for changes, test the output, and repeat this process many times. Even after launch, maintenance continues. Bugs appear, users request new features, and someone must verify that AI-generated changes are safe and correct. For more complex products, this ongoing labor remains a real cost.

Because of this, not every software product can become a healthy business. If customers can create a similar internal package with limited effort, the product may not be saleable at a sustainable price. The article suggests that viable software must offer enough value above the build threshold, the point where building it internally becomes reasonable. In practice, this means founders should look for products that are not only possible to build with AI, but still hard enough, broad enough, or costly enough to maintain that customers prefer to buy them.

The U.S. National Science Foundation, or NSF, is reportedly cutting some research programs to support a new technology initiative. According to insiders cited by Science, the agency is shifting money and attention away from parts of its existing portfolio. The NSF is a major public funder of science in the United States, so changes in its priorities can strongly affect universities, labs, and early-stage projects.

Supporters of the move may argue that a focused initiative can help the country respond faster to urgent technology needs. In government policy, an initiative is a new plan designed to reach a clear goal. It may support areas such as advanced computing, engineering tools, or other strategic technologies. A more concentrated investment can create momentum, but it can also reduce support for basic research, which often produces important discoveries over a longer period.

Critics are concerned about the process as well as the impact. When research programs are reduced quickly, scientists may lose grants, delay experiments, or change hiring plans. Universities also depend on stable funding to train graduate students and build teams. Insiders say the changes could weaken parts of the research pipeline, meaning the step-by-step path from early ideas to useful applications. If that pipeline becomes less stable, innovation may slow in the future.

The reported decision highlights a common policy challenge: how to balance immediate national goals with broad scientific support. Technology strategy often requires trade-offs, especially when budgets are limited. For engineers and researchers, the story is a reminder that funding choices shape which problems get solved. Even when a new initiative sounds promising, many experts say long-term progress depends on keeping a diverse portfolio of research programs across science and technology.