The application of TEE in AI agents

In the past few days, two AI agents have utilized a technology that has been around for a while but has not been very popular: TEE.

TEE stands for Trusted Execution Environment.

It is a hardware security-dependent solution.

It refers to an execution environment built through a combination of hardware and software within a computing device. This environment is typically used to protect sensitive data and perform critical operations, preventing unauthorized access and malicious actions.

In simple terms, if we run software in a reliable TEE environment, the outside world cannot crack it or obtain data and information during the execution process.

In the past, when this technology was mentioned, it was usually applied in environments like IoT devices and cloud computing. In the past two years, some encryption projects have also mentioned using this technology, such as wallets.

However, overall, I have always felt that it is aimed at the enterprise market, and there seems to be little potential for applications in personal consumer scenarios, so I basically did not pay much attention.

In the crypto ecosystem, there is a project using this technology called Phala Network. It provides a cloud computing platform based on TEE technology. In its white paper, it describes a way to provide this execution environment using distributed computing power.

I had seen this project earlier. I remember after reading its white paper, I thought of the Filecoin model, and thus felt that this so-called use of distributed computing power to achieve TEE was merely an "imagined" application, purely a forced fit with blockchain.

But recently, two new emerging AI agents, Spore (Spore.fun) and aiPool (@aipool_tee), have actually adopted the technology from this project.

In a previous article, when I introduced the currently widely implemented AI agents, I wrote that most of these AI agents require humans to apply for their crypto wallets. After humans apply, they then hand over the wallets for the AI agents to use.

In this case, since humans possess the wallet private keys, they can completely intervene in the operation of the AI agents—the simplest and most straightforward method is to directly transfer the assets out of the wallet.

Therefore, such AI agents cannot be considered "autonomous," at least not financially autonomous.

However, Spore and aiPool operate entirely within the TEE environment of the Phala network, and their wallets and keys are fully controlled by themselves, making it impossible for humans to manipulate their wallets or transfer their assets.

Thus, in terms of control over crypto assets, AI agents have now achieved complete autonomy over crypto assets through TEE technology, breaking free from human control.

In my view, this is truly an unexpected surprise of TEE technology applied in the field of AI agents.

If this path continues to develop, every fully autonomous AI agent will need to operate in a TEE environment and generate its own crypto wallet private keys, then the use cases for TEE technology will far exceed what I once imagined, primarily targeting traditional businesses and enterprise users, and will also find significant application space in the vast market of AI agents.

As for the Phala Network project itself, although I still believe that the model described in its white paper is somewhat forced, it has nonetheless found its application scenarios in these two typical cases. This is a remarkable event, as it has truly brought TEE technology into a promising crypto track.

Just a couple of days ago, I was lamenting in an article: sometimes the development of technology is truly unexpected; the usage scenario we envision is A, but the scenario that later flourishes is actually B.

TEE can be considered a case in point.

Back to AI agents, their recent development can be described as rapid. These developments are not just visions in my mind, but are gradually turning my previous imaginations into living reality.