AI 服務暫時不可用，以下為來源正文，待恢復後補全翻譯。
Linux is already part of the AI developer’s desktop toolkit. Nvidia’s DGX Spark, which became available to order on October 15, 2025, runs the Ubuntu-based DGX OS. OpenAI’s Codex CLI already supports Linux, another example of AI tools meeting developers where they work. There’s just one problem. Developers can fall in love with Linux at home. Getting enterprise IT to let them use it at work is another matter. Angelo Huang, founder and CEO of Swif.ai, a platform that manages and secures employee devices, wants to change that. Huang tells The New Stack that giving developers Linux machines starts with giving IT teams the controls to manage and secure them. Huang says that starting this past week, Swif began expanding its Linux device management and compliance capabilities, adding Omarchy and Linux Mint support to its roster. Swif is also optimizing the onboarding process across supported Mac, Windows, Linux, iOS, and Android environments. They claim that it now takes under 10 minutes to set up device compliance standards such as SOC 2, ISO 27001, and HIPAA. “Linux already runs much of the infrastructure behind AI,” Huang says. “Increasingly, developers want it on their desks too. The problem isn’t that Linux can’t be secure. It’s that enterprise IT needs to be able to manage it, enforce controls, and prove those controls are working for their compliance requirements. That’s the gap we’re closing.” “The problem isn’t that Linux can’t be secure. It’s that enterprise IT needs to be able to manage it, enforce controls, and prove those controls are working…” Swif lists support for Linux distributions including Debian, Ubuntu, Fedora, openSUSE, Arch Linux, Manjaro, MX Linux, Pop!_OS, Red Hat Enterprise Linux, NixOS, and Universal Blue. According to Swif’s Omarchy guide, the tested configuration (version 4.0.2) for x86_64 machines supports device enrollment, inventory tracking, application blocking, and selected OpenClaw logging and tool policies. (The guide flags limitations around login, home encryption, remote desktop, and updates.) Linux Mint’s status is less straightforward, though. “Linux Mint works today because it’s built on Ubuntu, and we have customers running it,” Huang says, “but it isn’t on our officially supported list yet. Requests for new distributions come in through support tickets and our customer success team. A customer with a real deployment, the number of seats involved, and how far it diverges from a distribution we already support move a distribution up the priority list. For an Ubuntu- or Fedora-based distribution, initial support typically takes one to two weeks.” Swif’s software-management guide lists Mint for APT-based application installation, although its Linux enrollment list does not name Mint. That documents a specific capability, rather than full validation of every management policy. Mint also offers a Debian-based edition, LMDE; shared ancestry alone does not establish compatibility across editions and versions. Making Linux manageable for enterprise IT Swif’s Linux management capabilities cover familiar IT needs: policy enforcement, LUKS encryption recovery, and automated remediation. Support varies by distribution and policy; the recovery policy manages access to an already encrypted disk rather than encrypting an unencrypted one. Swif sends Linux device evidence to the governance, risk, and compliance (GRC) platform Vanta hourly. It also offers a Drata integration with hourly syncing. Still, Swif’s Drata guide is inconsistent on Linux: it lists Linux evidence fields, but its device eligibility list names only macOS and Windows. Linux teams should confirm Drata coverage with Swif before relying on it. These records support compliance assessments; they do not, by themselves, establish compliance. The New Stack asks how a developer might, for example, run Wayland on a Framework laptop, and how this leaves IT to determine which management functions work with that setup. “Our agent runs in user space and relies mainly on systemd, the package manager, and standard kernel interfaces, so hardware like a Framework laptop is rarely the issue,” Huang says. “The display server is where it gets interesting. Wayland support through PipeWire and xdg-desktop-portal is in development. Our rule of thumb: if it affects whether our agent can enforce policy, report inventory, or prove compliance, it’s ours to fix.” User space does not mean unprivileged: Swif’s Linux management commands can run as root. PipeWire and desktop portals handle functions like screen sharing; this is distinct from basic device inventory or policy enforcement. Swif’s Omarchy guide flags its RustDesk-based remote desktop policy as having experimental Wayland support. Let the developers have Linux “The argument inside companies doesn’t need to be Mac versus Windows versus Linux,” says Huang. “If your AI team wants Linux, the answer from IT shouldn’t automatically be no. Give developers the tools they want to get the job done. Give security the controls it needs.” “If your AI team wants Linux, the answer from IT shouldn’t automatically be no.” Swif focuses on enterprise IT, but developers also have a stake in how IT manages their machines. The New Stack asked if it ran a Linux development community. “The usual story is that a few engineers ask IT to get off Windows, and it grows to 50, 100, sometimes 200. IT then needs a way to manage those machines that a GRC platform will accept as evidence, and that’s when they come to us. We also hear from developers directly, through shared Slack channels with customers, onboarding calls during rollouts, and support tickets filed by end users. We don’t run a community for Linux developers today, but it’s a good idea.” AI oversight: Documented features and roadmap “Right now,” Huang says, “we have MCP server management, agentic security, Claude Compliance integration, and Data loss prevention in beta. On the roadmap is integration with OpenAI’s compliance API and agentic compliance.” After enrolling a Linux machine, Swif’s MDM agent is installed for device management, while installing the Agentic Security agent provides visibility into supported AI-agent activity. Its Claude compliance integration collects organization-level audit events, while its documented DLP beta uses a browser extension and requires organization-level beta opt-in. Its email documentation covers Gmail and Outlook on the web, excluding native desktop mail clients. Those features have a different scope than monitoring local agent activity. “It’s bringing oversight of AI agents on employee endpoints into the device management layer. The problem is simple. Agents run on employee endpoints with access to code, credentials, and customer data, and almost nobody oversees them. “Agents run on employee endpoints with access to code, credentials, and customer data, and almost nobody oversees them.” “When OpenClaw first launched, many customers asked us to block it outright. As more people use agents at work, companies have too much to lose from a data breach or IP theft to leave that unmanaged,” Huang says. OpenClaw’s own security documentation warns that untrusted content can carry prompt injections. The risk depends on the agent’s permissions, tools, and isolation. The post A few engineers asked IT for Linux. Then it spread to 200 machines. appeared first on The New Stack.