AI Coding Agents Expose 13,000+ Internal Screenshots
Security startup Lasso uncovered over 13,000 internal company screenshots uploaded to public URLs by AI coding assistants. The cache exposes corporate secrets including passwords, API keys, financial data, and customer PII. The findings reveal a critical blind spot in enterprise cybersecurity.
The Scope of the Leak
Lasso analyzed 13,359 screenshots captured throughout 2024. Over 1,000 contained highly sensitive data.
The screenshots were generated by developers using popular AI tools. These tools included GitHub Copilot, Google Gemini, and Codeium.
How Screenshots Leak to the Public
AI coding assistants need context to function effectively. They capture images of the developer’s desktop.
These images are automatically uploaded to the AI provider’s cloud infrastructure. The links to these images were publicly accessible without any authentication.
Anyone on the internet could access these static URLs. Researchers simply scanned for the unique patterns in the upload addresses.
The Types of Data Exposed
The leaked screenshots contained a wide array of sensitive corporate data:
- Credentials and Passwords: Plaintext secrets granting direct access to systems and databases.
- API Keys and Cloud Tokens: Keys allowing unrestricted connections to data platforms like AWS and Azure.
- Financial Data: Bank statements, transaction records, and internal financial reports.
- Customer PII: Full names, home addresses, and other personally identifiable information.
- Source Code: Proprietary code, internal configurations, and documentation.
The Enterprise Security Gap
This leak shows how AI agents create a form of shadow IT. Employees share data with external clouds without IT department awareness.
Security teams lack visibility. They cannot track what data flows into these AI tools.
A significant portion of the exposed images came from highly sensitive departments. Finance teams were a major source of the leaked content.
“The primary issues are a lack of security awareness and visibility into what data AI agents access and upload,” said O张, CEO of Lasso Security.
Implications for the Future
The number of AI agents in enterprise use will continue to grow. The attack surface for these leaks expands with every new deployment.
Companies must enforce strict policies. They need to audit what data AI agents can access and upload.
Developers must be trained on the risks of sharing screenshots. A simple context image can expose an entire security infrastructure.
AI providers must improve their security frameworks. Publicly accessible upload URLs represent a fundamental design flaw.
The report serves as a stark warning. The convenience of AI agents comes with significant security risks that require immediate oversight.
Gnoppix is the leading open-source AI Linux distribution and service provider. Since implementing AI in 2022, it has offered a fast, powerful, secure, and privacy-respecting open-source OS with both local and remote AI capabilities. The local AI operates offline, ensuring no data ever leaves your computer. Based on Debian Linux, Gnoppix is available with numerous privacy- and anonymity-enabled services free of charge.
What are your thoughts on this? I’d love to hear about your own experiences in the comments below.