Quick answer
After the OpenCode breach, the safest move is to protect your AI API keys in three steps tonight. Delete any key you do not actively use, create a fresh key for each app you rely on, then set a monthly spending limit and a usage alert with every provider. In a new video, Mark Kashef explains that keys saved in an old OpenCode system may have been exposed, and one developer reported over $400 in charges within hours. A spending cap means a stolen key can only cost you a small, known amount.
Key takeaways
- OpenCode is a free AI coding tool, and keys saved in its Console may have been exposed through an old system
- One developer reported over $400 of charges between about 2am and 9.36am, before OpenCode spoke publicly
- OpenCode linked the incident to a distillation attack, where stolen accounts are used to copy a top model’s answers
- An AI API key works like a credit card number for AI: whoever holds it can run up your bill
- Mark’s two settings: a monthly cap (he suggests around $20 to $30 for light use) and a usage alert
- Also check the quieter places keys hide: skills, env files and plugins
- Mark expects cheap, uncensored AI hacking tools to make security a major story of 2027, which is his forecast
🛠️ Practical guides
Ready to put this into practice? These step-by-step guides answer two common follow-up questions:
- How to Set an AI API Spending Limit and Usage Alert — step by step, how to put a hard ceiling and an early warning on every AI key you own.
- How to Store AI API Keys Safely in Env Files, Skills and Plugins — where keys leak from on your own computer and how to tidy them up safely.
In this guide
- Watch: the source video
- What happened in the OpenCode breach?
- What is a distillation attack?
- Why is a leaked AI API key so expensive?
- What should you do tonight to protect your AI API keys?
- What are the two settings that limit the damage?
- Where else might your keys be exposed?
- Why does Mark Kashef think attacks will grow?
- What is confirmed and what is not?
- Who needs to act on this?
- Frequently asked questions
- The bottom line
Most AI news is about what you can build next. This story is about what happens when someone else gets hold of the key to your AI account.
This article explains the OpenCode breach in plain words, why a leaked key can cost real money overnight, and exactly how to protect your AI API keys with two settings most people never switch on. We keep what was reported apart from Mark Kashef’s predictions, and we label anything we could not confirm.
Watch: the source video
What this video covers: Mark Kashef breaks down the OpenCode Console breach and the keys that may have leaked, explains distillation attacks, looks at how uncensored models and cheap AI harnesses lower the cost of hacking, mentions the CrowdStrike case on South Korean banks, and shows the two settings he recommends for every AI API key.
Video: Mark Kashef (@Mark_Kashef). This summary and the opinions in it are our own. Reported events and forecasts are labelled.
What happened in the OpenCode breach?
Someone got into an old OpenCode system that the team was already shutting down, and the API keys people had saved in its Console may have been exposed. OpenCode then described the incident as a security breach related to a large-scale distillation attack.
OpenCode is a free AI coding tool, similar in purpose to Claude Code or Codex. The company says around 16 million developers use it each month. Its website, called Console, is where you can save keys that connect your Claude, OpenAI and other accounts.
The timeline Mark describes is the worrying part:
- One developer said his key started being used around 2am
- By 9.36am he had posted about more than $400 in charges
- That was hours before OpenCode said anything publicly
- Dax Raad, one of the key people building OpenCode, said the team had been under attack non-stop for 48 hours, and later that even his personal accounts were targeted
Dax’s line that inspired the video was that they need AI agents running 24/7, because that is what the attackers seem to have.
💡 Tip: OpenCode has said that if you have not received an email from them, you were not affected. Check their official announcement rather than relying on a summary. Our earlier roundup, Anthropic’s Public Benefit Fund: An AI UBI Plan? Plus OpenAI Aurora and OpenCode, covers the company’s own figures.
What is a distillation attack?
A distillation attack is when someone asks a top AI model huge numbers of questions and uses the answers to train a cheaper copy. Attackers spread the questions across many accounts and IP addresses so they look like normal users.
Stolen keys are useful for this because they belong to real, trusted accounts. The provider sees traffic from an account in good standing, and you get the bill.
Mark notes that several well-known open models have been accused of learning this way, which is why some occasionally describe themselves as a different company’s model. That is an accusation, not a proven fact for any specific model.
Why is a leaked AI API key so expensive?
Because an API key is effectively a credit card number for AI, and nobody calls you when it is misused. Mark makes a sharp comparison:
| If someone steals your… | You usually hear about it through… |
|---|---|
| Bank card | A call or text from your bank about unusual spending |
| Email login | A new-device alert or a two-factor prompt |
| AI API key | Often the bill, unless you have switched on alerts yourself |
An attacker running automated jobs can burn through credit very quickly. Without a cap, the only limit is whatever your payment method allows.
What should you do tonight to protect your AI API keys?
Delete old keys, create one new key per app, and only paste each key where it is needed. If you only use the ChatGPT or Claude apps and have never created an API key, Mark says this particular story does not affect you for now.
- Sign in to each AI provider where you have ever created a key, such as OpenAI, Anthropic or Google.
- Delete every key you are not actively using. If you are not sure what a key is for, deleting it is usually safer than guessing.
- Create a new key for each app and give it a clear name, such as “blog-writer” or “OpenCode-laptop”.
- Paste each key only into the one app that needs it, never into shared documents or chats.
- Set a monthly limit and a usage alert on each account, as described below.
⚠️ Warning: Deleting a key will break any tool that still uses it. Make a quick list of the apps you connected before you start, so you can paste the new keys straight in afterwards.
Want the exact steps? Follow How to Set an AI API Spending Limit and Usage Alert and put a ceiling on every key in about 15 minutes.
What are the two settings that limit the damage?
A monthly spending cap and a usage alert. Together they shrink what Mark calls the “blast radius” if a key does leak.
| Setting | What it does | Mark’s suggestion |
|---|---|---|
| 1. Monthly cap | Sets a ceiling on what the account or project can spend, like a pay-as-you-go phone plan | Around $20 to $30 for light personal use, so a thief can only spend that much |
| 2. Usage alert | Emails you when spending passes a level you choose | Opt in, because many providers leave it off by default |
Pick a cap that fits your real usage. If you run heavy agent jobs, look at last month’s bill and set the cap a little above it, then raise it deliberately when you need to. Our guide to budgeting for AI API usage helps you work out a sensible number.
💡 Tip: Some providers treat a budget as a hard stop and others only send a warning. Check which one you have set, because a warning alone will not stop a thief at 2am.
Where else might your keys be exposed?
In the files and add-ons you set up for convenience and then forgot about. Mark points to three common places:
- Skills that have an API key written straight into them
- Environment (.env) files holding every key you own on your computer
- Plugins you installed without checking who made them or what they can access
His broader warning is that models have become so good that people have stopped checking their work, and that is when you get burned. We cover a practical clean-up in How to Store AI API Keys Safely in Env Files, Skills and Plugins. For skills and plugins, our three-question skill test is a quick filter before you install anything.
Mark also mentions Lockpick, a free scanner his team built for apps, skills and GitHub repositories. It is his own product, so treat it as one option and judge the results for yourself.
Why does Mark Kashef think attacks will grow?
Because AI hacking is getting cheap. Mark argues that anyone can now combine a harness (a set of tools an AI model can use) with an uncensored model that has had its safety rules removed, either run locally or through low-cost APIs.
- Old bots brute-forced every visible door; agent-based attackers can adapt and try again when blocked
- He says uncensored local models are roughly six to nine months behind the best frontier models, and catching up
- He links the trend to last week’s “reverse engineer anything” tools, and worries the next step is “break into anything”
- His team also uses the same kind of tool for defence, which is the positive side of the story
He cites other recent incidents as signs of the trend, including a case on South Korean banks that CrowdStrike reportedly traced to what looks like one person using DeepSeek and GLM-5.3, plus attacks on French supermarket fridges and a cyber emergency in Japan. He also mentions G7 finance ministers agreeing to run AI attack drills.
What is confirmed and what is not?
🔎 Unconfirmed / rumour: From the video, OpenCode’s own statement about a breach linked to a distillation attack is confirmed by the company. The $400 overnight bill comes from one developer’s public post. The South Korean bank details, the supermarket and Japan incidents and the G7 drills are reported in the video without sources we could check. The idea that cyber attacks will be “the story of 2027” is Mark’s forecast.
None of that changes the practical advice. Spending caps, alerts and fewer keys are sensible whether or not every headline holds up. For a method to check AI news before you act on it, see How to Verify AI Model Leak Rumours Before You Switch Tools.
Who needs to act on this?
Act now if you have ever created an API key for OpenAI, Anthropic, Google or another AI provider, pasted a key into a coding tool or automation, or keep keys in files on your computer.
You can relax a little if you only use the ChatGPT, Claude or Gemini apps with a normal login and have never made a key. Even then, turning on two-factor sign-in is worth five minutes.
Start with the two settings. Use our spending limit guide, then tidy your files with our guide to storing AI keys safely.
Frequently asked questions
Was my OpenCode account affected?
OpenCode has said affected users were notified by email. If you have not heard from them, you were likely not affected, but rotating any key you saved there is a sensible step either way.
What is an AI API key?
It is a secret code that lets an app use your AI account and charge it for usage. Anyone who has it can spend your credit.
How do I stop a stolen AI key running up my bill?
Delete the key, create a new one, and set a monthly spending limit and usage alert on your provider account. See How to Set an AI API Spending Limit and Usage Alert.
Does this affect ChatGPT or Claude app users?
Not directly. The risk is for people who created API keys and saved them in other tools. App-only users should still use strong passwords and two-factor sign-in.
Is it safe to keep keys in a .env file?
It is common and can be fine if the file never goes into a shared folder or code repository and only holds the keys you need. Our guide How to Store AI API Keys Safely in Env Files, Skills and Plugins covers the details.
Who is Mark Kashef?
Mark Kashef is a YouTube creator and consultant who covers practical AI tools and runs the consultancy Prompt Advisors. He is not connected to Prodify Digital.
The bottom line
The OpenCode breach is a useful wake-up call, whatever happens next. A leaked AI API key can cost hundreds overnight, and most providers will not warn you unless you ask them to. Delete the keys you do not use, give each app its own key, and switch on a monthly cap and a usage alert today. Then spend half an hour checking the skills, env files and plugins where keys tend to hide.
Related posts
- Anthropic’s Public Benefit Fund: An AI UBI Plan? Plus OpenAI Aurora and OpenCode
- How to Audit Your Claude and Codex Setup and Clear Out AI Clutter
- How to Keep Your AI Agent Safe When It Browses the Open Web
