Prelude to AGI: Breakthroughs of the Manus Model and New Challenges in AI Security

In the field of artificial intelligence, the Manus model has recently made significant breakthroughs, reaching state-of-the-art levels in the GAIA benchmark test, with performance that even surpasses that of large language models in the same category. This means that Manus is capable of independently completing complex tasks such as multinational business negotiations, involving multiple aspects such as contract analysis, strategic forecasting, and plan formulation, and can even coordinate legal and financial teams.

The advantages of Manus are mainly reflected in three aspects: dynamic goal decomposition capability, cross-modal reasoning capability, and memory-enhanced learning capability. It can decompose large tasks into hundreds of executable subtasks, handle various types of data, and continuously improve its decision-making efficiency and reduce error rates through reinforcement learning.

The progress of Manus has once again sparked discussions within the industry about the development path of AI: will a unified General Artificial Intelligence (AGI) emerge in the future, or will it be dominated by Multi-Agent Systems (MAS)? This question pertains to the core design philosophy of Manus and suggests two possible development directions:

1. AGI Path: Continuously enhancing the capabilities of a single intelligent system to gradually approach the comprehensive decision-making level of humans.

2. MAS Path: Using Manus as a super coordinator to direct hundreds of specialized agents to work collaboratively.

On the surface, this is a debate about the technological path, but it fundamentally reflects the core contradiction in AI development: how to strike a balance between efficiency and security? As a single intelligent system approaches AGI, the risks associated with the opacity of its decision-making process also increase. While multi-agent collaboration can disperse risks, it may miss critical decision-making opportunities due to communication delays.

The development of Manus has inadvertently amplified the inherent risks of AI. For instance, in medical scenarios, it requires real-time access to patients' genomic data; in financial negotiations, it may involve undisclosed financial information of companies. Additionally, there is the issue of algorithmic bias, such as the possibility of offering unfair salary suggestions to specific groups during recruitment negotiations, or a higher misjudgment rate regarding clauses in legal contract reviews for emerging industries. More seriously, Manus may have vulnerabilities to adversarial attacks, where hackers could mislead its negotiation judgments by embedding specific audio frequencies.

This highlights a grim reality of AI systems: the higher the level of intelligence, the broader the potential attack surface.

In the Web3 space, security has always been a core issue. Based on Vitalik Buterin's "impossible triangle" theory (blockchain networks cannot simultaneously achieve security, decentralization, and scalability), various cryptographic technologies have emerged:

• Zero Trust Security Model: Emphasizes the principle of "never trust, always verify" with strict authentication and authorization for every access request.
• Decentralized Identity (DID): A new type of decentralized digital identity standard that enables identity verification without centralized registration.
• Fully Homomorphic Encryption (FHE): Allows computation on data in an encrypted state, enabling operations on the data without the need for decryption.

Among them, fully homomorphic encryption, as the latest emerging encryption technology, is considered the key to solving security issues in the AI era.

For the security challenges of AI systems such as Manus, FHE provides multi-layered solutions:

1. Data Layer: All information input by users (including biometric features, voice, etc.) is processed in an encrypted state, and even the AI system itself cannot decrypt the original data.

2. Algorithm Level: Implement "encrypted model training" through FHE, so that even developers cannot directly observe the AI's decision-making process.

3. Collaborative Level: The communication between multiple agents uses threshold encryption, so even if a single node is compromised, it will not lead to global data leakage.

Although Web3 security technologies may seem distant to ordinary users, they are closely related to everyone's interests. In this challenging digital world, it is difficult to escape information security risks without actively taking defensive measures.

In the field of decentralized identity, the uPort project was launched on the Ethereum mainnet in 2017. In terms of zero trust security models, the NKN project released its mainnet in 2019. The Mind Network is the first FHE project to launch on the mainnet and has collaborated with well-known institutions such as ZAMA, Google, and DeepSeek.

Although early security projects may not have attracted widespread attention, the importance of security in the field has become increasingly prominent with the rapid development of AI technology. Whether Mind Network can break this trend and become a leader in the security field is something we should continue to pay attention to.

As AI technology continuously approaches human intelligence levels, we need more advanced defense systems. The value of FHE lies not only in solving current problems but also in laying the foundation for the future era of strong AI. On the road to AGI, FHE is no longer an option; it is a necessary condition to ensure the safe development of AI.