When AI Meets Fully Homomorphic Encryption: An Igniting Open-Source Movement

AI is deeply penetrating into core fields such as healthcare, finance, and government affairs, but data privacy remains an insurmountable obstacle. Hospitals are reluctant to hand over medical records to cloud models, banks dare not let third - parties access transaction records, and enterprises are hesitant to use external AI to analyze internal data. It's not that the technology is inadequate; rather, the cost of trust is too high.

Fully Homomorphic Encryption (FHE) is changing this situation. It enables AI to perform inferences directly on encrypted data - the model cannot see the original data, and the data provider cannot see the model parameters, yet the calculation results are completely accurate. It sounds like magic, but it's real cryptography. Moreover, it is transitioning from theory to engineering.

CipherFlow Launches the "Prometheus Project"

Recently, CipherFlow officially launched the LattiAI open - source co - construction plan, the "Prometheus Project", targeting global developers, university researchers, and AI engineers to jointly refine LattiAI, the first domestic fully homomorphic encryption development framework for AI ciphertext inference.

As the kick - off event of the plan, CipherFlow will host an online workshop at 19:00 on April 1st (Wednesday). Four experts from Peking University, Tsinghua University, and CipherFlow will be invited to share insights on the current status and future directions of FHE + AI technology:

• Talk 1: Privacy Crisis in the AI Era and the Breakthrough Path of FHE —— Fan Junfeng, CipherFlow
• Talk 2: When AI Meets Cryptography: A Paradigm Revolution in Progress —— Li Meng, Peking University
• Talk 3: On the Eve of Explosion: Research Opportunities and Directions of FHE —— Wang Anyu, Tsinghua University
• Talk 4: Introduction to the Open - Source Project latti - ai and Co - construction Guide —— Chen Si, CipherFlow

Three Tracks to Cover Participants with Different Backgrounds

The first phase of the "Prometheus Project" sets up three tracks:

In this event, we have set up three tracks to cover participants with different backgrounds: Track 1: AI + FHE Framework Capability Building Suitable for: Deep learning engineers and framework developers.

• Task 1: Implement ciphertext conversion of a classic CNN model, such as VGG, Inception, SSD
• Task 2: Implement homomorphic computation of the softmax layer

This track is targeted at deep learning engineers. Participants do not need a cryptography background. The framework has encapsulated the FHE encryption and decryption processes, allowing participants to focus on model structure adaptation and inference implementation.

Track 2: Implementation of Secure AI in Industry Scenarios

Suitable for: Researchers and engineers in financial/medical AI.

• Task 3: Implement ciphertext inference in the financial field, such as price prediction and credit assessment models
• Task 4: Implement ciphertext inference for a model in the medical and health field, such as electrocardiogram analysis and medical imaging models

This track is for researchers and engineers with industry model experience. Participants will adapt existing industry AI models to ciphertext inference solutions based on the latti - ai framework to verify the engineering feasibility of FHE in real - world scenarios.

Track 3: FHE Underlying Optimization

Suitable for: Cryptography researchers and those in the high - performance computing field.

• Task 5: Optimize computational efficiency using sparse - packed bootstrapping operators
• Task 6: Optimize FHE parameters for low - precision neural networks

This track is for researchers in the fields of cryptography and high - performance computing, requiring a foundation in FHE algorithms.

Time Nodes

• Workshop: 19:00 - 20:30 on April 1st (Wednesday), 2026
• Task claim deadline: 19:00 on April 15th, 2026
• Code + PPT submission deadline: 19:00 on April 30th, 2026
• Review and award ceremony: Online announcement in May + Offline Meetup

Award Settings

• Outstanding Contribution Award: ¥10,000 × 1
• Excellent Sharing Award: ¥5,000 × 2
• Ecosystem Contribution Award: ¥3,000 per team or individual

Outstanding results will be integrated into the latti - ai open - source framework and become part of the project ecosystem. The expert advisory team comes from universities such as Peking University, Fudan University, Beihang University, and Nanjing University of Aeronautics and Astronautics.

For more details about the event, follow the official WeChat account of CipherFlow.