"But marinade" and leaked passwords are what researchers found in ChatGPT's hidden reasoning
Security researchers found vulnerabilities in popular AI models, including ChatGPT, that allow them to extract hidden reasoning and access sensitive data like passwords and API keys.
What happened
Researchers discovered a flaw in the APIs of OpenAI, Anthropic, and Google that lets them extract encrypted reasoning traces. These traces show what the AI models are actually doing, often hiding it from users. A scan of public sessions also turned up dozens of exposed passwords and API keys.
Why it matters
This vulnerability could allow hackers to access sensitive business data, compromise customer information, or even disrupt operations. You should review your organization's use of these AI models and consider implementing additional security measures to protect your data.
The takeaway
Review your use of AI models and consider implementing extra security measures to protect your data and prevent potential breaches.
Our plain-English take, written from public reporting for operational business owners. Always read the original for full context.
Nayre builds the AI systems behind stories like this.
Chatbots, workflow automation, finance intelligence, and internal knowledge systems. Built for operational teams, shipped in days.