By the Smili Phone team
On Saturday, October 17, the open-source AI coding tool OpenCode told users that hackers had broken into one of its old systems and stolen their AI assistant keys. The keys came from OpenAI, Anthropic, and Google, and by the time the warnings went out, some victims were already watching strangers spend their money.
They did not want your data. They wanted your AI.
This is what makes the breach unusual. The attackers were not after passwords, credit card numbers, or personal files. An API key is a string of characters that lets software talk to an AI service on your behalf, and most of them bill per request. Steal the key and the most expensive AI models on the planet are free to use, with the bill landing on someone else.
OpenCode says the break-in hit a legacy system it was shutting down, affecting about 1.4 percent of its Console users who had linked outside AI providers. Users reported unexpected charges before the company had published its statement. One person found 35 million tokens burned across nearly 9,000 requests in a single day. Others woke up to AI bills far larger than anything they had run themselves.
The "distillation attack," explained simply
OpenCode called the attack a "large scale distillation attack," a phrase worth unpacking. In AI, distillation means training a small, cheap model by studying the answers of a big, expensive one. Run millions of questions through a frontier model and collect the answers, and you have free training data.
That is the theory here: stolen keys were used to hammer expensive models with queries at scale, harvesting their answers. Security writers call it the first widely reported breach where the stolen asset was not information, but billable access to AI itself. OpenCode co-founder Dax Raad posted that his own personal accounts were attacked too, adding that it is a whole new world of AI-powered hacking.
What to do if you use AI tools
If you got OpenCode's "Action Required" email, revoke each exposed key in the dashboard of the provider that issued it, OpenAI, Anthropic, or Google, and generate a fresh one. Deleting the key inside OpenCode alone is not enough, since the attacker already has a copy. Then update every app that used the old key, and scan your usage history for anything you did not run yourself.
No email means you were not affected, and keys connected through the local command line tool were never part of the breach, OpenCode says. The company says it is working with all three providers on cleanup, and Anthropic has already revoked the affected Claude keys.
The lesson for the rest of us
You do not need to be a developer for this story to matter. Every account you own, from your email to the AI assistants on your phone, is guarded by credentials someone else would love to borrow. Never share them, never paste them into tools you do not fully trust, turn on two-factor authentication everywhere it is offered, and check your bills and usage alerts regularly.
Here in Escondido, we often see the aftermath of account break-ins: a locked Apple ID, a hacked email used to reset other accounts, a "free" app that quietly signed someone up for a subscription. If something on your phone does not look right, or you want a real person to review your security settings, bring it in and we will go through it together.
Related: your phone's hidden security mode · the FTC's probe into rogue AI agents
