InfoSecBulletin Cybersecurity for mankind
Security researchers found that datasets used by companies to develop large language models included API keys, passwords, and other sensitive credentials.
Large language models are dominating the online landscape, with companies promoting AI solutions that claim to solve all problems.
CVE-2026-20349, CVE-2026-68820
Cisco and Windows patched zero days exploited in attack
Microsoft Patch 394 Flaws, Including 3 Zero-Days
AI Finds Nation-State-Level Zoom Flaw in Under 24 Hours, Zero-Click Attack Exposed
Hacker Actively Exploit SonicWall and SharePoint Flaws
Gunra Ransomware Leverage Fortinet VPN Flaws to Evade MFA Obtaining Enterprise Data
Hackers accessed a US defense manufacturer’s Microsoft 365 account via phishing.
Google Play Apps Utilize Stealth Loaders to Spread Anatsa Banking Malware
Bangladeshi org listed “the Gentlemen’ ransomware victim list: CIRT warns
Around 800 Malicious npm Packages Distribute Cross-Platform RAT and Infostealer
Google Chrome 151 Update Fixes 41 Flaws, 6 Critical
For an AI to be effective, it needs extensive training data, much of which is gathered from the Internet by specialized companies and organizations.
Common Crawl provides datasets for companies to train their AI, gathering information from the internet, which may include sensitive data.
Researchers from Truffle Security discovered that credentials, API keys, and passwords are being exposed. The main issue is that some web developers hardcode sensitive information into websites, which then ends up in LLM training data.
Researchers discovered 11,908 live secrets, such as API keys and passwords, across 2.76 million websites.
“Leaked keys in Common Crawl’s dataset should not reflect poorly on their organization; it’s not their fault developers hardcode keys in front-end HTML and JavaScript on web pages they don’t control. And Common Crawl should not be tasked with redacting secrets; their goal is to provide a free, public dataset based on the public Internet for organizations like Truffle Security to conduct this type of research,” explained the researchers.
Companies that create LLMs have warned against hardcoding sensitive information on websites. They advise avoiding this practice, as users may unintentionally share the code in their work, worsening the issue.
