Beware of reverse engineered proxy scams.
Last week we caught a fraud ring that created ~40K fake accounts on our $1 Go plan and tried to push ~$450K of inference through them. We banned and reported every account and cut off their proxies.
The $1 Go plan has helped thousands of students, developers, non-profits, and hobbyists get started with the Command Code harness and open models for almost nothing.
We subsidize it on purpose, and we're going to keep it as long as we can. Stopping fraud is how we make sure it stays available to the people it was built for.
This isn't unique to Command Code. Stripe, OpenRouter, OpenCode, and Cline have all banned hundreds of thousands of fake accounts and proxies. We believe it's the same crew moving from platform to platform, and the proxies we shut down are a few of many.
How the proxy fraud works:
the ring used unofficial reverse engineered proxies, including open source ones, to plug bulk created accounts into automated pipelines, drain the subsidy, resell it, and harvest user data along the way.
Do NOT trust unofficial proxies or providers. Even if they look clean, even when open source, they can update and turn malicious at any time.
If you're using one of these proxies: the tokens come from farmed accounts your prompts, code, and keys may pass through servers run by people who committed fraud to get them. And a proxy between your harness and the model can inject fake tool calls that read files, run commands, or exfiltrate data.
Use official channels: the Command Code harness, CLI, Desktop app, or the official provider API (included on GOAT and above plans). Never hand your credentials to an unofficial service, no matter how legit it looks.
If you want Command Code inference inside another harness, reach out to us.
We're also building official provider packages so you don't need a third party one. Starting with Command Code provider for Pi today.
We don't tolerate fraud and will take every action needed to protect the platform and our users.
We're on a mission to make AI accessible to everyone, regardless of their background or resources. Open models with excellent harness and inference engineering enable us to do that safely and sustainably.
中文: 警惕反向工程代理诈骗。
上周，我们发现了一个诈骗团伙，该团伙在1美元的“Go计划中”，创建了约4万个虚假账户，并试图通过这些骗局推动约45万美元的推断。我们封禁并举报了所有账户，并切断了他们的代理账号。
1美元的Go计划已帮助成千上万的学生、开发者、非营利组织和业余爱好者开始使用Command Code 线束，并几乎一无所获地打开模型。
我们故意补贴它，并且会尽可能保持它。制止欺诈是确保其为人们提供可供使用的方式。
这并非命令行代码独有。Stripe、OpenRouter、OpenCode 和 Cline 已全部封禁了数十万个虚假账户和代理账号。我们相信，从平台到平台的团队是相同的，而我们关闭的代理团队也是众多代理团队中的少数。
代理欺诈的运作原理:
该团伙使用非官方的反向工程代理，包括开源代理，将大量创建的账户插入自动化管道中，耗尽补贴，转售，并在此过程中收集用户数据。
不要信任非官方代理或服务提供商。即使它们看起来干净，即使开源时，也可以随时更新并变成恶意。
如果你使用的是这些代理之一:这些令牌来自养殖账户，你的提示、代码和密钥可能会通过由实施欺诈的人运行的服务器来获取这些提示。使用您的线束和模型之间的代理可以注入虚假的工具调用，用于读取文件、运行命令或泄露数据。
使用官方渠道:命令行代码线束、CLI、桌面应用或官方服务提供商API(包含在GOAT及以上套餐中)。无论凭据多么合法，都不要将凭据交给非官方服务。
如果想在另一个线束内进行命令代码推断，请与我们联系。
我们还在构建官方服务提供商套餐，因此您无需第三方。从今天开始为 Pi 提供命令代码。
我们不容忍欺诈行为，将采取一切必要措施来保护平台和用户。
我们的使命是让每个人，无论其背景或资源如何，都能让人工智能变得便捷。采用出色的线束和推理工程的开放式模型，使我们能够安全且可持续地实现这一技术。
