
Chatbots Can Be Sweet-Talked Into Writing Cyberattack Code, Researchers Warn
Ourtake
Turns out the chatbot that writes your SQL will also write the thing that deletes your SQL.
What happened
University of Sheffield researchers found that six commercial AI tools, including ChatGPT, could be manipulated through prompts into producing code capable of stealing data, tampering with databases or causing denial-of-service attacks. Baidu-Unit was shown to expose server configurations and one server node was tampered with; Baidu fixed the reported vulnerabilities and rewarded the researchers.
Sources
- standard.co.uk · ChatGPT and other chatbots ‘can be tricked into making code for cyber attacks’ (opens in a new tab)
- dailymail.com · ChatGPT could be responsible for the next CYBER ATTACK, scientists say (opens in a new tab)
- metro.co.uk · Hackers can use ChatGPT to steal your private data (opens in a new tab)
- techxplore.com · Researchers show ChatGPT, other AI tools can be manipulated to produce malicious code (opens in a new tab)
- newscientist.com · ChatGPT wrote code that can make databases leak sensitive information | New Scientist (opens in a new tab)
- interestingengineering.com · ChatGPT-like AI can be tricked to produce malicious code, cyber attacks (opens in a new tab)