Original author: Defi 0x Jeff, head of steak studio
Original translation: zhouzhou, BlockBeats
Editors Note: This article focuses on a variety of technologies that enhance privacy and security, including zero-knowledge proof (ZKP), trusted execution environment (TEE), fully homomorphic encryption (FHE), etc., and introduces the application of these technologies in AI and data processing, how to protect user privacy, prevent data leakage, and improve system security. The article also mentions some cases, such as Earnifi, Opacity, and MindV, showing how to achieve risk-free voting, data encryption processing, etc. through these technologies, but at the same time these technologies face many challenges, such as computing overhead and latency issues.
The following is the original content (for easier reading and understanding, the original content has been reorganized):
As the supply and demand for data surges, the digital footprint left by individuals becomes more extensive, making personal information more vulnerable to misuse or unauthorized access. We have already seen some cases of personal data breaches, such as the Cambridge Analytica scandal.
For those who haven’t caught up yet, check out Part 1 of the series, where we discussed:
Importance of data
The growth of data demand from artificial intelligence
The emergence of the data layer
Europe’s GDPR, California’s CCPA, and regulations elsewhere around the world have made data privacy not just an ethical issue but a legal requirement, pushing companies to ensure data protection.
As the development of artificial intelligence has proliferated, AI has also further complicated the areas of privacy and verifiability while improving privacy protection. For example, while AI can help detect fraudulent activities, it also enables deep fake technology, making it more difficult to verify the authenticity of digital content.
advantage
Privacy-preserving machine learning: Federated learning allows AI models to be trained directly on the device without centralizing sensitive data, thereby protecting user privacy.
AI can be used to anonymize or pseudonymize data so that it cannot be easily traced back to an individual, while still being useful for analysis.
AI will be critical to developing tools to detect and mitigate the spread of deepfakes, thereby ensuring the verifiability of digital content (and detecting/verifying the authenticity of AI agents).
AI can automatically ensure that data processing practices comply with legal standards, making the verification process more scalable.
challenge
AI systems often require large data sets to work effectively, but how that data is used, stored, and accessed can be opaque, raising privacy concerns.
With enough data and advanced AI techniques, it is possible to re-identify individuals from supposedly anonymous datasets, undermining privacy protections.
As AI is able to generate highly realistic text, images, or videos, it becomes more difficult to distinguish between real and AI-forged content, challenging verifiability.
AI models can be deceived or manipulated (adversarial attacks), undermining the verifiability of data or the integrity of the AI system itself (as seen in the cases of Freysa, Jailbreak, etc.).
These challenges have driven the rapid development of AI, blockchain, verifiability, and privacy technologies, leveraging the strengths of each technology. We have seen the rise of the following technologies:
Zero-knowledge proof (ZKP)
Zero-knowledge transport layer security (zkTLS)
Trusted Execution Environment (TEE)
Fully Homomorphic Encryption (FHE)
1. Zero-Knowledge Proof (ZKP)
ZKPs allow one party to prove to another party that they know certain information or that a statement is true, without revealing any information beyond the proof itself. AI can use this to prove that data processing or decisions meet certain criteria without revealing the data itself. A good case study is getgrass io, which uses idle Internet bandwidth to collect and organize public web data for training AI models.
Grass Network allows users to contribute their idle Internet bandwidth through browser extensions or applications, which is used to crawl public web data and then process it into structured data sets suitable for AI training. The network performs this web crawling process through nodes run by users.
Grass Network emphasizes user privacy and only captures public data, not personal information. It uses zero-knowledge proofs to verify and protect the integrity and origin of data, prevent data corruption, and ensure transparency. All transactions from data collection to processing are managed through sovereign data aggregation on the Solana blockchain.
Another good case study is ZKME.
zkMe’s zkKYC solution addresses the challenge of conducting the KYC (Know Your Customer) process in a privacy-preserving manner. By leveraging zero-knowledge proofs, zkKYC enables platforms to verify user identities without exposing sensitive personal information, thereby protecting user privacy while maintaining regulatory compliance.
2. zkTLS
TLS = a standard security protocol that provides privacy and data integrity between two communicating applications (often associated with the “s” in HTTPS). zk + TLS = improved privacy and security in data transmission.
A good case study is OpacityNetwork.
Opacity uses zkTLS to provide a secure and private data storage solution. By integrating zkTLS, Opacity ensures that data transmission between users and storage servers remains confidential and tamper-proof, thereby solving the privacy issues inherent in traditional cloud storage services.
Use Case — Earnifi, an app that has reportedly climbed to the top of the app store rankings, particularly in the financial category, leverages OpacityNetwork’s zkTLS.
Privacy: Users can provide lenders or other services with their income or employment status without revealing sensitive banking information or personal data, such as bank statements.
Security: The use of zkTLS ensures that these transactions are secure, verified, and remain private. It avoids the need for users to trust their entire financial data to a third party.
Efficiency: The system reduces the costs and complexity associated with traditional wage advance access platforms, which may require cumbersome verification processes or data sharing.
3.TEE
Trusted Execution Environments (TEEs) provide hardware-enforced isolation between the normal and secure execution environments. This is probably the most well-known security implementation in AI agents right now to ensure they are fully autonomous agents. Popularized by 123skelys aipool tee experiment: a TEE presale where the community sends funds to the agent, which issues tokens autonomously according to pre-defined rules.
PhalaNetwork by marvin tong: MEV protection, ElizaOS with ai16z dao integration, and Agent Kira as a verifiable autonomous AI agent.
fleeks one-click TEE deployment: Focuses on simplifying usage and improving developer accessibility.
4. FHE (Fully Homomorphic Encryption)
A form of encryption that allows computations to be performed directly on encrypted data without first decrypting the data.
A good case study is mindnetwork xyz and their proprietary FHE technology/use cases.
Use Case — FHE Heavy Staking Layer and Risk-Free Voting
FHE Heavy Collateral Layer By using FHE, heavily collateralized assets remain encrypted, which means private keys are never exposed, significantly reducing security risks. This ensures privacy while also verifying transactions.
Risk-free voting (MindV)
Governance voting is conducted on encrypted data, ensuring that voting remains private and secure, reducing the risk of coercion or bribery. Users gain voting power by holding heavily staked assets (vFHE), thus decoupling governance from direct asset exposure.
FHE + TEE
By combining TEE and FHE, they create a strong security layer for AI processing:
TEE protects operations in the computing environment from external threats.
FHE ensures that operations are always performed on encrypted data throughout the process.
For institutions processing $100M to $1B+ in transactions, privacy and security are critical to prevent front-running, hacking, or exposure of trading strategies.
For AI agents, this double encryption enhances privacy and security, making it useful in the following areas:
Sensitive training data privacy
Protect internal model weights (to prevent reverse engineering/IP theft)
User data protection
The main challenge of FHE remains the high overhead due to its computational intensity, resulting in increased energy consumption and latency. Current research is exploring methods such as hardware acceleration, hybrid cryptography, and algorithm optimization to reduce the computational burden and improve efficiency. Therefore, FHE is best suited for low-computation, high-latency applications.
Summarize
FHE = operates on encrypted data without decrypting it (strongest privacy protection, but most expensive)
TEE = hardware, secure execution in an isolated environment (balance between security and performance)
ZKP = proves a statement or authenticates an identity without revealing the underlying data (good for proving facts/credentials)
This is a broad topic, so this is not the end. A key question remains: in an age of increasingly sophisticated deepfakes, how can we ensure that AI-driven verifiability mechanisms are truly trustworthy? In Part III, we’ll dive deeper into:
Verifiability layer
The role of AI in verifying data integrity
The future of privacy and security
