OpenCode Got Hacked (What Can We Learn From It)

Imagine that. Imagine that just getting spending alerts of thousands and you just don't know where it came from. Like that would be my worst nightmare. Command AI Open code, one of the best, most inclusive open-source coding harnesses out there, got targeted today by some pretty, I guess, malicious, sickening people. Uh, we got this message here directly from Open Code. We had a security breach last night related to a large-scale distillation attack that impacted 1.5% of Open Code console users. [snorts] We have notified them all this morning via email. If you did not get an email, you are not impacted. It was a subset of users who had external providers linked. The incident is ongoing, so we cannot share more details at this point, but we are working with Anthropic, OpenAI, and Google to remediate. While the company didn't specify how the distillation attack occurred, basically leaked API keys. They found a bunch of API keys to all of these, um, these providers. And overnight developers began posting spending alerts and unauthorized charges. I think I saw multiple thousands of, uh, dollars being spent on people's API keys. Cuz again, Open Code has its own links to to use the, uh, to different models. But you if you've got, say, you want to use Claude, you can put in your API key and it will use Claude will take that money from you instead of Open Code. So, this is unclear, unproven, but that's what they think they were trying to do, basically. Like I say, they they got like given people were getting alerts. Imagine that. Imagine that just getting spending alerts of thousands and you just don't know where it came from. Like that would be my worst nightmare. OpenAI said it'd be working with Open Code to restore access and refund API customers. I would I'm guessing they're making a note of that here that Anthropic and Google aren't looking to do that, but we'll we'll see. We'll see. The API keys have been blocked, so they can't use them. This is a really good faith thing from OpenAI, I think, to to be refunding people. The open source app and command line tool were not affected, so this is the thing that you download. Again, it's this open code console thing that they talk about. No, this is the open code console, not open code itself. Those keys never come to us. They had to use agents to keep working for 48 hours, requiring the team to keep AI defensive agents running around the clock. So, it's no longer about humans jumping on and whatever. You They're literally getting their agents to like patch things, block people, look at logs, and all the rest of it. I mean, it makes perfect sense, but it's just crazy. 48 hours of doing that. Cuz the thing is, and this is what happened with with Hugging Face, is Hugging Face weren't able to use Claude or OpenAI because they've got guardrails in place to prevent people from doing these attacks. Yeah. So, what do they call it? They call it like the asymmetry problem. They call it the asymmetry problem, where it's like, you know, you're kind of Pretty devastating attack, and and I think we're going to be seeing more and more of this, but you know, common cybersecurity practices is and I know we don't do this, but you really need to think carefully about the API keys, where they're stored. You know, they're a password at the end of the day, but a lot of these tools allow you to do two things. They allow you to set expiry dates on these things, and I've started doing this on my tutorials because in in case a API key leaks on one of my tutorials, which it has done in the [clears throat] past, it I I just set a day expiry, so by the time the video's gone out, that API key's expired. You can put credit limits on these things. So, and this is less about API keys, more about the the platforms themselves, but you can spend you can put spending limits. These This is what the professionals do. They really, I think, by default, GitHub doesn't let you put a non-expiring personal access token. They they make you interesting put 90 days. So it's just this is ongoing personal pri- uh security practices in in full swing. Like you should be looking after your apps and they shouldn't have unlimited access, unlimited spending. They should have an expiry. You should be readdressing your security every few months, you know? Uh if you're looking after an app, it's like, "Okay, cool. Are we still hashing things? Do we need to rotate keys?" Maintaining protection is part of what it takes to run a commercialized application. It It's crazy. You don't notice these when you vibe code something, but it will hit you once you scale or like once a breach like this happens. I did put up a project, a really stupid game, and it hit uh 1.5 million CDN requests in like 4 days. I did actually check if uh Vercel is going to automatically charge me if something like this hap- I didn't anticipate this, but I was just like I was just careful. I was like, "I don't know how much the hobby allows me to do because I'm on the free plan. So if I hit the upper limit, will it automatically upgrade and you know, send me like a $2,000 bill or will it limited or like downgraded or something like this?" I disabled uh my app, everything that I had, and I redirected everything temporarily to the launch page for the waitlist. And I have like now 1,500 emails. [laughter] Nice. Nice. Nice. It It wasn't all for nothing. This is part of a larger conversation on my show Command AI, which we stream live every single week. We discuss the news and all things related to AI in the world of design and web. Catch us next week and join in the banter. Command AI