.webp?w=1600&fit=1600,900&ssl=1 "EncGPT Encryption & Decryption Rules")
Researchers from Xi’an Jiaotong University have introduced EncGPT, an AI-powered multi-agent framework that dynamically generates encryption and decryption rules.
This innovation addresses critical challenges in communication security, balancing cost-efficiency and high-level encryption reliability. EncGPT leverages large language models (LLMs) to create a dynamic encryption workflow that adapts to evolving security needs, marking a significant leap forward in cryptographic technology.
Traditional encryption methods often struggle with inherent limitations. Symmetric encryption, while fast, risks compromising security if keys are intercepted, whereas asymmetric encryption offers higher security but at the cost of increased computational complexity. Moreover, static algorithms are vulnerable to advanced cyberattacks due to their predictability.
EncGPT’s Multi-Agent Workflow
Dynamic encryption technologies aim to overcome these issues by introducing adaptability into cryptographic systems. EncGPT takes this concept further by dynamically generating encryption rules and algorithms instead of relying on fixed methods.
EncGPT operates through three specialized agents: the Rule Agent, Encryption Agent, and Decryption Agent.
- Rule Agent: This agent generates encryption rules dynamically during each communication round. It uses classical algorithms like Caesar and Vigenère ciphers as foundational templates, modifying them with randomly generated keys and parameters to ensure unpredictability.
- Encryption Agent: It applies the rules generated by the Rule Agent to encrypt user inputs.
- Decryption Agent: This agent decrypts ciphertext using the same dynamic rules, ensuring seamless communication flow.
The architecture ensures that plaintext only appears during user input and output stages, enhancing security by preventing “prompt contamination,” a phenomenon where previous encryption rules interfere with subsequent interactions.
Researchers tested EncGPT using GPT-4o as the core engine for rule generation and encryption-decryption workflows. The system demonstrated strong performance with substitution algorithms like Caesar, Vigenère, and Atbash ciphers.
| Algorithm | Rule Gen (s) | Encryption (s) | Decryption (s) | Total (s) |
|---|---|---|---|---|
| Caesar | 13.36 | 6.45 | 7.51 | 27.23 |
| Vigenère | 14.13 | 8.42 | 8.34 | 30.89 |
| Atbash | 10.44 | 6.57 | 7.32 | 24.33 |
Homomorphic properties allowing operations on encrypted data without decryption were successfully preserved during tests. The average execution time for these workflows was 15.99 seconds, highlighting efficiency but also room for optimization compared to traditional heuristic algorithms.
EncGPT addresses several key issues in cryptography:
- Enhanced Security: By dynamically generating encryption algorithms and keys, EncGPT eliminates vulnerabilities associated with static methods.
- Cost Efficiency: Classical algorithms modified dynamically are computationally simpler and less resource-intensive.
- Homomorphic Encryption: EncGPT supports operations on encrypted data without exposing sensitive information—a feature crucial for privacy-preserving applications like financial data processing.
EncGPT represents a significant advancement in communication encryption by introducing an agent-based workflow that dynamically generates encryption algorithms and keys.
This approach enhances encryption reliability and addresses communication security challenges in LLM-MA systems.
Investigate Real-World Malicious Links & Phishing Attacks With Threat Intelligence Lookup - Try for Free
