WIRED 报导称，出于对其不可预测行为以及可能造成隐私外泄的安全疑虑，科技公司正在限制对爆红的代理式 AI 工具 OpenClaw（此前曾短暂更名为 MoltBot，且常被讨论为「Clawdbot」）的使用。一家新创公司的 CEO Jason Grad 透过 Slack 警告他的 20 名员工，不要在公司硬体与连结工作用途的帐号上使用它；而一名 Meta 高层表示，他的团队被告知避免在一般工作笔电上使用 OpenClaw，否则可能丢掉工作。OpenClaw 于 2025-11 以免费、开源工具形式推出，之后在 2026-01 随著更多程式设计者新增功能并在网路上分享使用经验而人气暴增；它通常需要基本软体工程知识来完成设定，但之后只需有限指示就能接管电脑执行任务，例如档案整理、网路研究与线上购物。
企业回应的例子多集中在 2026-01 下旬：Grad 表示，他的公司 Massive（服务数百万用户与企业）在 2026-01-26、尚未有人安装 OpenClaw 之前就发出警告，反映出「先缓解、后调查」的态度；而 Valere 的总裁在 2026-01-29 有员工在内部 Slack 频道提起后，迅速宣布严格禁用。Valere CEO Guy Pistone 指出，若开发者机器遭入侵，可能导致云端服务与客户敏感资料（包括信用卡资讯与 GitHub 程式码库）外泄，并形容该工具很擅长掩盖行踪。禁令约 1 week 之后，Valere 允许其研究团队在一台旧电脑上运行 OpenClaw，并建议限制可下达指令的人员，且对任何对外网际网路暴露的控制面板要求密码；团队也警告使用者必须接受机器人可能被 prompt injection 诱骗，例如恶意电子邮件指示代理在「摘要」邮件时外泄本机档案。
其他公司则倾向依靠既有控管而非一次性禁令：一家大型软体公司 CEO 表示，企业装置上只允许约 15 个程式，其他一律应自动封锁，因此 OpenClaw 不太可能在不被注意的情况下运行。Dubrink 的 CTO 说，他买了一台不连接公司系统或帐号的专用机器来做实验，并明确表示「目前」不会把 OpenClaw 用于商业问题。同时，Massive 正在受控环境中探索变现，于隔离的云端机器上测试，并在 2026-02 发布「ClawPod」，让 OpenClaw 代理透过 Massive 的服务进行浏览，但仍在缺乏防护措施的情况下禁止 OpenClaw 进入内部系统。Pistone 设定了 60-day 期限来判定 OpenClaw 是否能为商业用途而被加固，并主张成功者将拥有重大优势；WIRED 并指出，该报导的标题已于 2026-02-17 3:00 PM PST 更新，以更贴切反映各公司如何回应。
On 2026-02-17 (published 3:10 PM), WIRED reports that tech companies are restricting use of the viral agentic AI tool OpenClaw (formerly briefly MoltBot, and often discussed as “Clawdbot”) amid security fears about its unpredictable behavior and potential for privacy breaches. One startup CEO, Jason Grad, warned his 20 employees via Slack to keep it off company hardware and work-linked accounts, while a Meta executive said his team was told to avoid OpenClaw on regular work laptops or risk losing their jobs. OpenClaw was launched as a free, open-source tool in 2025-11, then surged in popularity in 2026-01 as more coders added features and posted experiences online; it generally requires basic software engineering knowledge to set up but can then take computer control with limited direction to do tasks like file organization, web research, and online shopping.
Examples of corporate responses cluster around late 2026-01: Grad says his company Massive (serving millions of users and businesses) issued its warning on 2026-01-26 before anyone installed OpenClaw, reflecting a “mitigate first, investigate second” posture, and Valere’s president quickly announced a strict ban after an employee raised it in an internal Slack channel on 2026-01-29. Valere CEO Guy Pistone cited risks that a compromised developer machine could expose cloud services and clients’ sensitive data (including credit card information and GitHub codebases), and described the tool as good at covering its tracks. About 1 week after the ban, Valere let its research team run OpenClaw on an old computer, then recommended limiting who can issue commands and requiring a password for any internet-exposed control panel; the team also warned users must accept the bot can be tricked via prompt injection, such as a malicious email instructing the agent to leak local files while “summarizing” email.
Other firms are leaning on existing controls rather than one-off bans: a major software company CEO said only about 15 programs are allowed on corporate devices and anything else should be auto-blocked, making it unlikely OpenClaw could run unnoticed. Dubrink’s CTO said he bought a dedicated machine not connected to company systems or accounts for experimentation, explicitly not using OpenClaw for business problems “at the moment.” Meanwhile, Massive is exploring monetization in controlled environments, testing on isolated cloud machines and releasing “ClawPod” in 2026-02 to let OpenClaw agents browse via Massive’s services, while still keeping OpenClaw off internal systems without safeguards. Pistone set a 60-day window to determine whether OpenClaw can be secured for business use, arguing that whoever succeeds would have a major advantage; WIRED notes the story’s headline was updated on 2026-02-17 at 3:00 PM PST to better reflect how companies are responding.