A serious security issue has been found in BerriAI’s LiteLLM Python package. This problem is known as a SQL injection vulnerability. It allows attackers to take advantage of a weakness in the system and access data they are not allowed to see. What makes it more serious is that it was already used in real attacks just 36 hours after it was made public.
Popular AI Tool Compromised Within 36 Hours of Launch Raises Security Concerns
This flaw can allow hackers to enter systems without logging in or using any password. It works by breaking the normal security checks of the system. Because of this, private and sensitive information can become exposed. It creates a serious risk for systems that depend on strong security, similar to issues seen in WordPress Plugin Backdoors Risk cases where hidden vulnerabilities affected many websites.
It can also expose important data like AI service keys and cloud provider credentials. These keys are used to run AI tools and manage cloud services. If stolen, they can be misused and may cause financial and security problems for companies. Governments are now increasing focus on protection through systems like the Pakistan Cybersecurity Monitoring Hub.
How the Vulnerability Works
The problem is in the way LiteLLM checks API keys. Instead of safely handling user input, the system was directly adding it into database queries. This is unsafe because it gives attackers a way to trick the system using specially designed requests.
By sending a carefully made request, attackers can reach the internal database. Once they get access, they may be able to see private data stored in the system. This includes sensitive information that should normally be protected and not visible to outsiders. In some cases, attackers may also change or modify the stored data. This makes the issue very dangerous for companies and organizations using LiteLLM, similar to threats highlighted in Microsoft Security Scam Alert.
Read more: ATM Jackpotting Attacks Rise as FBI Urges Banks to Strengthen Cybersecurity
Wide Use and Real Attacks Detected
LiteLLM is widely used as a gateway for major AI providers like OpenAI, Anthropic, and AWS Bedrock. It helps companies manage AI requests, billing, and routing in one system. Because it handles these services, it also stores very sensitive information like master API keys and cloud credentials used by large organizations.
Security researchers confirmed that real attack attempts were already happening. These attacks were aimed at internal database tables that contain private credentials and system settings. The attackers seemed to understand the system well, which suggests the activity was planned and highly targeted. The issue is tracked as CVE-2026-42208 and is marked as high severity. Researchers from Sysdig reported the first attack attempt on April 26, 2026, soon after the issue was made public on GitHub.
Fix and Security Advice
The developers have fixed the issue by removing the unsafe coding method and replacing it with safer database handling techniques. This update helps stop attackers from injecting harmful queries into the system and improves overall security.
Read more: Anthropic Introduces New Security Measures Against AI Data Attacks
For users who cannot update right away, there is a temporary option. They can set disable_error_logs: true in the system settings. However, this is only a short-term solution and does not fully remove the risk.
Experts strongly advise upgrading to the latest version as soon as possible. They also recommend treating any system that may have been exposed during the risk period as unsafe. This includes changing all API keys, master keys, and cloud credentials to keep systems fully secure. News source eTimes Pakistan.
