I previously mentioned in several articles that AI Agents would be the 'redemption' of many old narratives in the crypto industry. In the last wave of narratives surrounding AI autonomy, TEE was once lifted to the forefront, yet there is a less popular technical concept, FHE—homomorphic encryption, which may also gain 'rebirth' due to the momentum of the AI track. Below, I will clarify the logic through examples.

FHE is a cryptographic technology that allows direct computation on encrypted data and is regarded as the 'Holy Grail.' Compared to popular technologies like ZKP and TEE, it is relatively niche, primarily constrained by overhead and application scenarios.

Mind Network focuses on FHE infrastructure and has launched the FHE Chain MindChain, which is dedicated to AI Agents. Despite raising over ten million dollars and spending years on technological development, market attention remains underestimated due to the limitations of FHE itself.

Recently, however, Mind Network has launched several positive announcements regarding AI application scenarios. For example, its developed FHE Rust SDK has been integrated into the open-source large model DeepSeek, becoming a critical part of AI training scenarios and providing a secure foundation for trustworthy AI. Can FHE perform in AI privacy computing and leverage the narrative of AI Agents for a breakthrough or redemption?

In simple terms: FHE homomorphic encryption is a cryptographic technology that can be directly applied to the current public chain architecture, allowing arbitrary computations such as addition and multiplication on encrypted data without prior decryption.

In other words, the application of FHE technology allows data to be fully encrypted from input to output. Even nodes that maintain public chain consensus for verification cannot access plaintext information. This enables FHE to provide a technical underpinning for training AI LLMs in vertical segments such as healthcare and finance.

This enables FHE to become a 'preferred' solution for enriching and expanding traditional AI large model training in vertical scenarios while integrating blockchain distributed architecture. Whether it is cross-institutional collaboration on medical data or privacy reasoning in financial transaction scenarios, FHE can serve as a supplementary choice due to its uniqueness.

This is not abstract; a simple example makes it clear: for instance, an AI Agent aimed at the C-end typically integrates AI large models from various suppliers like DeepSeek, Claude, and OpenAI in its backend. But how can we ensure that in some highly sensitive financial application scenarios, the execution process of the AI Agent is not suddenly influenced by a large model backend that changes the rules? This certainly requires encrypting the input prompts so that when LLM service providers process the ciphertext directly, there will be no forced interference that affects fairness.

So what is the concept of 'trustworthy AI'? Trustworthy AI is a decentralized AI vision that Mind Network attempts to build with FHE, allowing multiple parties to achieve efficient model training and inference through distributed computing power without relying on central servers, providing consensus verification based on FHE for AI Agents. This design eliminates the limitations of centralized AI and provides dual guarantees of privacy and autonomy for web3 AI Agents operating in a distributed architecture.

This aligns more closely with the narrative direction of Mind Network's distributed public chain architecture. For example, during special on-chain transactions, FHE can protect the privacy reasoning and execution processes of all parties' Oracle data, allowing AI Agents to make autonomous trading decisions without exposing positions or strategies, etc.

So, why is it said that FHE will have a similar industry penetration path as TEE, bringing direct opportunities due to the explosion of AI application scenarios?

Previously, TEE captured the opportunity of AI Agents thanks to its hardware environment, which allows data to be managed in a privacy-preserving state, enabling AI Agents to autonomously manage private keys and achieve a new narrative of asset management. However, there is a critical flaw in TEE's management of private keys: trust relies on third-party hardware providers (e.g., Intel). For TEE to function effectively, a distributed chain architecture is needed to impose an additional transparent 'consensus' constraint on TEE environments. In contrast, PHE can exist entirely based on a decentralized chain architecture without relying on third parties.

FHE and TEE have similar ecological niches. Although TEE's application in the web3 ecosystem is not yet widespread, it is already a mature technology in the web2 field. In contrast, FHE is expected to gradually find its value in both web2 and web3 amid the current AI trend explosion.

That’s all.

In summary, it is evident that FHE, as a cryptographic holy grail-level technology, is bound to become one of the foundational stones for security as AI becomes a prerequisite for the future, with the likelihood of further widespread adoption.

However, despite this, it is essential to address the cost issue associated with the implementation of FHE algorithms. If it can be applied in web2 AI scenarios and then linked to web3 AI scenarios, it is likely to unexpectedly release a 'scale effect' that dilutes overall costs, allowing for more widespread application.