📰 English IT Daily · 2026-06-14

A public message called “Open source AI must win” argues that artificial intelligence should not be controlled by only a small number of closed companies. The writer says AI is becoming basic infrastructure for work, education, science, software, creativity, and public services. If people can only rent intelligence through closed services, they may lose more than software freedom. They may also lose the practical freedom to use and manage important systems on their own terms.

The message warns about dependence on closed APIs, remote platforms, changing terms of service, unclear moderation rules, limited model access, and prices set by a few firms. An API is a tool that lets one software service connect to another. If a company relies fully on a remote AI service, it may be affected when costs rise, features change, or access is restricted. This can create what the writer calls a kind of subscription economy for cognition, where thinking tools are rented instead of owned or controlled.

In contrast, the statement says open source AI should stay usable, understandable, reproducible, locally deployable, economically viable, and community-governed. In simple terms, this means people should be able to study how systems work, rebuild them, run them on their own machines, and improve them without asking permission. The writer also says AI systems should be easy to inspect, modify, benchmark, teach, and preserve over time, even if major labs, cloud platforms, or hardware vendors change direction or disappear.

The wider argument is about long-term capacity and independence. The message says intelligence infrastructure is too important to depend on a few private gatekeepers. It supports strong national capacity together with global open standards, meaning shared technical rules that many groups can use. For businesses, developers, and public institutions, this debate is not only philosophical. It affects cost control, security review, system reliability, and the ability to adapt AI tools to local needs in the future.

A report said talks between Amazon’s chief executive and U.S. officials were followed by tighter controls on some AI models from Anthropic. The case has drawn attention because it connects business discussions, government concerns, and access to powerful AI systems. Anthropic is known for building large language models, which are AI systems trained on huge amounts of text to understand and produce language.

The reported crackdown appears to be related to national security and the risk that advanced models could be misused. In simple terms, a crackdown means stronger action to limit or control something. Governments around the world are trying to decide how to manage frontier models, a term often used for the most advanced AI systems. These models can help with coding, research, and productivity, but they may also create new risks if safety rules are weak.

Amazon has invested in Anthropic, so any change in model access can affect business strategy as well as public policy debates. When companies work closely with AI partners, decisions about model availability, safety checks, and compliance can become very important. Compliance means following laws, rules, and internal standards. For cloud and software teams, this issue is not only political; it can also change which tools are available for developers and enterprise customers.

The report highlights a larger question for the AI industry: who should decide how powerful models are released and used? Some people argue that governments need stronger oversight to protect society. Others worry that unclear rules could slow innovation and reduce competition. For engineers and business leaders, the key lesson is that AI development now depends not only on technical performance, but also on trust, regulation, and responsible deployment.

A recent blog post explained how one developer built a local coding agent on macOS after several internet failures left him without access to online AI tools. He wanted a system that was fast enough to use on a Mac, worked through an OpenAI-compatible API, and could also handle images such as screenshots. In the end, he created a setup that could respond at a usable speed on an Apple M1 Max machine with 64 GB of unified memory.

The final setup used llama.cpp with Metal acceleration, a version of Gemma 4 in GGUF format, a draft model for speculative decoding, a multimodal projector, and Pi as the terminal coding agent. In simple terms, Metal acceleration lets the model use Apple hardware more efficiently, while GGUF is a file format designed for running large language models locally. A multimodal system can process both text and images, which is useful when a developer wants to show the agent what appears on the screen.

The main model alone produced about 58 tokens per second during generation. The blog said this speed was usable, but not ideal for coding-agent work, especially when the agent makes many tool calls. The developer then added an MTP draft model. MTP, or multi-token prediction, helps the system guess several next tokens in advance. This is part of speculative decoding, a method that can increase output speed without changing the main model itself.

After testing several settings, the best result on that machine was about 72 tokens per second, which was around 24% faster than the main model alone. Prompt processing stayed almost the same, while generation became clearly faster. The writer also noted that the best setting depends on the hardware, so users should test different values on their own systems. The article shows that a local AI coding agent on macOS is now practical for some developers, especially when reliability, privacy, and offline access are important.

Security researchers have reported a strange spyware case linked to malicious open-source packages. According to a post that shared research from Socket, some malware developers added text about nuclear and biological weapons into their code or package materials. Researchers say this kind of text may have been used as a distraction, a joke, or a way to make analysis harder. The key issue, however, is not the wording itself but the spyware behavior hidden inside the packages.

The reported campaign appears to target bioinformatics and MCP developers. Bioinformatics is a field that uses software to study biological data, while MCP usually refers to tools that help software systems connect with outside services. In attacks like this, developers may install a package that looks useful but actually includes malicious code. Once installed, spyware can collect information from a device, watch user activity, or send data to an outside server without clear permission.

This case shows a growing risk in the software supply chain. The software supply chain includes the code, libraries, and packages that developers depend on every day. If one package is poisoned, many users can be affected. Open-source software is important and widely used, but it also requires careful checking. Researchers often advise teams to review package sources, check unusual behavior, and limit what third-party code can access.

For companies, the lesson is practical. Security teams should monitor package use, train developers to spot warning signs, and respond quickly when a threat is reported. Clear policies for dependency management can reduce risk. This means knowing which packages are approved, who added them, and whether they have changed suddenly. Even when attackers use shocking or unusual text, organizations need to stay focused on the real problem: hidden spyware inside trusted development tools.

Azure Compute has announced the retirement of several virtual machine series for Azure Batch pools. The affected series are Av2, F, Fs, Fsv2, G, Gs, and Lsv2. The retirement date is November 15, 2028. A virtual machine, or VM, is a software-based computer in the cloud, and a Batch pool is a group of VMs used to run many jobs at the same time.

This change matters to teams that use these VM series for large-scale processing. Azure Batch is often used for tasks such as data processing, rendering, testing, and other workloads that can be divided into many smaller jobs. After the retirement date, new Batch pools using these VM series will no longer be available. Existing users should review their current pool configurations and understand where these older VM types are still in use.

A retirement notice does not always mean an immediate problem, but it gives organizations time to plan. In cloud services, retirement means a product or option will stop being supported or offered after a certain date. For engineers, this usually leads to a migration project. Migration means moving workloads from one VM series to another that can provide similar performance, storage, or memory for the application.

For business and technical teams, early planning is important. They may need to test replacement VM series, check application compatibility, compare cost, and update automation scripts or deployment templates. Even though 2028 is still some years away, this kind of announcement can affect long-term architecture decisions. By preparing early, teams can reduce risk, avoid service disruption, and make a smoother transition for future Batch workloads.

vocabulary

Azure Compute has announced that several older virtual machine series will be retired on May 1, 2028. The change will affect Azure Batch pools that use these machine types. The list includes D-series, Ds-series, Dv2-series, Dsv2-series, and Ls-series. A virtual machine, or VM, is a software-based computer in the cloud. A Batch pool is a group of VMs used to run many jobs automatically, such as large data or computing tasks.

According to the update, after the retirement date, new Batch pools using these VM series cannot be created. Existing pools that rely on them will also be affected, so teams should review their current setup well before the deadline. Retirement means a product or service will no longer be available for normal use. In cloud operations, this can create risk if systems continue to depend on older resources without a migration plan.

For engineering teams, the announcement is a reminder to check infrastructure regularly and avoid waiting until the last minute. Batch workloads often support important business processes, testing, analytics, or background processing. If a pool uses a retired VM series, jobs may need to move to a newer option. Migration usually means selecting another supported VM size, testing performance, and confirming that applications, storage, and cost expectations still match business needs.

Although the update does not provide every technical detail in the source context, the main message is clear: organizations using these VM series in Batch pools should start planning early. Good preparation can reduce service disruption and help teams make better architecture decisions. This kind of notice also shows why cloud environments need ongoing review. Even stable systems can change over time, and support policies can have a direct impact on operations.