How We Taught ChatGPT-4 to Break mbedTLS AES With Side-Channel Attacks

ChatGPT-4 used to break mbedTLS AES encryption with power measurements, demonstrating a highly effective side-channel attack that reduces attack time and effort, with insights on the new TOK4 system and its limitations.

Key takeaways
  • ChatGPT-4 can be used to break mbedTLS AES encryption with side-channel attacks, specifically using power measurements.
  • The attack works by analyzing the power consumption of the device during encryption and deciphering the key.
  • The attack can be performed in a matter of seconds and is highly effective, reducing the cost and complexity of side-channel attacks.
  • The use of ChatGPT-4 in side-channel attacks can significantly reduce the time and effort required to break encryption.
  • The presentation did not disclose the specific plugin used for the attack, but emphasized its ease of use and effectiveness.
  • The authors proposed a new side-channel attack system using ChatGPT-4, which they call TOK4, which reduces the token size by a factor of 4.
  • The system uses advanced encryption standard (AES) encryption with wolfSSL library and can break the encryption in a matter of seconds.
  • The authors demonstrated a live demo of the attack and showed how ChatGPT-4 can be used to break encryption with side-channel attacks.
  • The authors emphasized the importance of secure communication and the need for robust encryption and countermeasures against side-channel attacks.
  • The presentation also touched on the limitations of ChatGPT-4, such as its inability to do math and its tendency to hallucinate, but did not elaborate much on these limitations.