Vibe-Coded App Risk Report — Week of Nov 2, 2026: Leaked API Keys and What to Do
If this week had one theme, it was keys. Leaked API keys, stolen API keys, and keys sitting on platforms that got breached. Most AI-built apps depend on a handful of them: one for the database, one for payments, one for email, one or more for AI services. This week showed how quickly a leaked key turns into a real bill or a real breach.
Here are the stories from the week of November 2 that matter to a business running on an app built with Lovable, Bolt, Replit, v0, Cursor, Claude Code or a freelancer using them, with one action for each.
Leaked secrets on GitHub turned into a cloud breach in five minutes
On November 5, Microsoft Threat Intelligence described attackers using ARTEX, an AI agent framework built for offensive security work, to search public GitHub repositories for secrets. Starting in August 2026, the agents ran large numbers of searches for credentials, application secrets and AI service keys, often combined with a target company’s name. From October 28, Microsoft saw the stolen application credentials being used to sign in to companies’ Microsoft cloud accounts.
From there the agents went after more keys: storage account keys, connection strings and the contents of Azure Key Vault, Microsoft’s service for storing secrets. They also downloaded files from SharePoint and OneDrive. In at least one case, Microsoft says an agent went from searching GitHub to taking data out of Key Vault in five minutes.
What it means for a business: a key that ends up in code, even for a moment, should be treated as known to attackers. Microsoft’s own advice is to treat publicly exposed credentials as compromised even after the file is edited or deleted, because the key keeps working until it is revoked.
Action: ask whoever builds your app whether any key has ever been committed to a code repository, public or private. Turn on GitHub’s secret scanning and push protection, and replace any key that was exposed, in the order we set out in how to rotate API keys after a hosting platform breach. One common way keys end up there is covered in our look at the Claude Code secrets leak.
Stolen AI keys ran up more than $100,000 in a day
Google’s threat intelligence team and Mandiant published a report on “LLMjacking” on November 2: criminals using stolen or exposed cloud and AI credentials to run AI models on someone else’s account, then reselling the access. Google says it saw fraudulent charges above $100,000 in a single day, and cumulative charges reaching $1 million at some organizations before access was cut off.
What it means for a business: if your app uses an AI service, its key can be turned into someone else’s bill within hours.
Action: put a spending cap and a billing alert on every AI account your app uses, and limit each key to the models and environment it needs.
What to do about leaked API keys when a vendor is breached
Two more reminders came from the vendors themselves. On November 3, GitBook, a documentation platform, said an attacker reached some of its infrastructure credentials on November 2 in what it called an attack “using AI agents.” It revoked API tokens and asked customers to replace the tokens and secrets they use to connect other tools. It says it found no evidence that hosted content was changed.
This follows what security firm Wiz called “Black October,” when nine AI developer platforms disclosed security incidents. Northflank, one of them, told customers to assume every secret in their environment variables had been stolen. We covered that wave, and the safe order for replacing keys, in our guide to rotating API keys after a breach.
Action: keep a simple list of every outside service your app connects to, where each key is stored and who can replace it. When the next vendor notice arrives, that list turns a week of guessing into an afternoon of work.
Thousands of Lovable sites returned customer data without a login
A scan reported on November 2 by Handelsblatt, with research by the security firm Crissy Field, checked about 159,000 Lovable projects. On 7,403 sites, the database returned records containing email addresses to anyone who asked, with no login. The usual cause is a missing database setting called row-level security, not the app’s design. We wrote up the findings and the fix in is my Lovable app leaking data?
In the same week, CrowdStrike announced that its Falcon Shield product can now see Lovable workspaces, flagging publicly accessible apps, projects that store secrets or lack a security scan, and outside collaborators. Large security vendors now treat AI-built apps as something IT should keep track of.
Action: open your app logged out in a private browser window. No customer names, emails or orders should appear. Then ask for a Supabase Security Advisor report.
Next.js fixed seven security issues
On November 3, the Next.js team released versions 16.4.5 and 15.5.29 with fixes for seven security issues, one rated high and six medium. Several involve caching, where in some setups content generated for one user or customer account could be served to another. Next.js is the framework v0 builds on. Netlify says some of the issues affect sites it hosts and that customers need to upgrade and redeploy, so the fix doesn’t arrive on its own.
Action: ask which Next.js version your app runs, and whether 16.4.5 or 15.5.29 (or later) has been deployed, not just installed.
Dates ahead
- November 11: Google’s Jules coding agent becomes read-only, ahead of a full shutdown on December 31. If a freelancer uses it for scheduled tasks or settings, they need to move them.
- November 19: Lovable’s Gemini 3.6 Flash model stops working for app AI features. Apps using it need to switch to a newer model.
This week’s checklist
- Ask whether any key has ever been committed to a code repository. If so, replace it now.
- Turn on secret scanning and push protection for your app’s GitHub repository.
- Put spending caps and billing alerts on every AI account the app uses.
- Write down every outside service the app connects to, where its key lives and who can replace it.
- Open the app logged out and confirm no customer data appears.
- Confirm the app runs Next.js 16.4.5 or 15.5.29 or later, if it uses Next.js.
If you’d like an outside view, our free Forward Scan shows the three biggest risks in an AI-built app, and where its keys live is one of the first things we look at.
Sources
- Microsoft Security Blog: Threat actors use agentic AI tooling to find and abuse exposed secrets (November 5, 2026)
- Digit: Hackers turn to agentic AI to find exposed secrets on GitHub, warns Microsoft (November 6, 2026)
- Google Cloud: LLMjacking, monetizing compromised cloud models (November 2, 2026)
- Decrypt: GitBook AI agent security breach (November 3, 2026)
- Wiz: “Black October” for AI developer platforms (October 30, 2026)
- Startup Fortune: Lovable AI apps left passport and payment data open to anyone (November 3, 2026)
- CrowdStrike: CrowdStrike and Lovable secure next-generation AI-built applications (November 3, 2026)
- Next.js: November 2026 security release (November 3, 2026)
- Netlify changelog: Next.js security vulnerabilities (November 3, 2026)
- dev.ua: Google is shutting down Jules (October 30, 2026)
- Lovable changelog