Private API keys and passwords found in AI training dataset - nearly 12,000 details leaked
Private API keys and passwords found in AI training dataset - nearly 12,000 details leaked
Security researchers discovered thousands of valid API keys and passwords within the Common Crawl dataset, a massive archive of web data. These hardcoded secrets, often left by developers during coding, were found across billions of pages. This exposes significant vulnerabilities, as malicious actors can easily exploit these unremoved credentials for unauthorized access. The findings are particularly alarming because major large language models rely on Common Crawl for training data. Despite existing filters intended to remove sensitive information, many secrets likely persist through the ingestion process. This suggests that generative AI systems may inadvertently learn and reproduce private data, posing severe privacy and security risks to users whose information was archived. This article is crucial for the open_data community as it highlights the inherent risks of publishing raw, unfiltered web archives. It underscores the necessity for robust data sanitization and privacy-preserving techniques before datasets are made publicly available or used for AI training. Developers and organizations contributing to open datasets must prioritize security to prevent the accidental leakage of sensitive information into public repositories and downstream technologies.
Source: techradar.comPublished on 2025-03-04