Future financial technology controlled by AI robot using machine learning and artificial … [+] intelligence to analyze business data and give advice on investment and trading decision. 3D illustration.

getty

AI agents—autonomous systems designed to make decisions, perform tasks, and interact within digital environments—are increasingly seen as transformative for various industries, including finance. These agents operate independently, following pre-set goals or adapting dynamically, and hold promise in roles ranging from customer service to fund management.

Investor interest in AI agent startups has surged recently: in the last 12 months, there have been 156 deals in the AI agent space, marking an 81.4% increase year-over-year, according to PitchBook. So far in 2024, AI agents alone have raised over $1 billion in funding, per CB Insights. Across the AI sector as a whole, investment levels are reaching historic highs, with $18.9 billion raised in Q3 2024 alone, comprising 28% of all venture funding—highlighted by OpenAI’s unprecedented $6.6 billion round, the largest venture deal of all time, according to Crunchbase.

As the potential of autonomous agents becomes more tangible, crypto is emerging as a promising infrastructure to enable AI agents to securely and independently manage funds, potentially overcoming the limitations of traditional finance systems.

AI agents annual funding rate.

CB Insights
The Problem with Traditional Finance for Autonomous AI Agents

In conventional finance, regulations like Know Your Customer (KYC) and Anti-Money Laundering (AML) laws are critical to ensure transparency, accountability, and ethical use of funds. These regulations, however, assume that a human is responsible for any financial account and has passed relevant identity and background checks. But in the case of AI agents, no single individual or legal entity may actually control the account directly, creating regulatory gray areas.

Moreover, financial tools and protocols in traditional banking are designed to serve human users. From online banking systems to investment accounts, each financial service is built on the assumption that there’s an accountable, legally recognized human or corporate entity behind every transaction. An AI agent operating independently doesn’t easily fit into these frameworks, making compliance both technically challenging and legally uncertain. Thus, for AI-driven finance to work on a practical level, a solution that sidesteps the limitations of traditional finance while addressing security and regulatory concerns is necessary.

“The credit card system simply isn’t built for machine-to-machine payments,” said Idan Levin, co-founder and CPO of OpenCommerce. “It’s riddled with inefficiencies, high transaction fees, and privacy compliance issues, which makes it unsuitable for autonomous agents. Stablecoins are a far better alternative, especially as they bypass many of these pain points, allowing for smoother, more reliable transactions.”

The KYC process for banks involves three main stages: customer identification, customer due … [+] diligence, and ongoing monitoring.

KYC Hub
Why Crypto Wallets Are an Ideal Solution for AI Agents

Crypto wallets are a compelling solution to the challenges of autonomous money management by AI. Unlike traditional banking accounts, which often require personal identification and human intermediaries, crypto wallets can be created and managed by software without direct human involvement. This independence makes crypto wallets an attractive choice for AI agents that need to manage funds autonomously.

Crypto wallets are also highly programmable. An AI agent can interact with a wallet’s API, setting rules for transactions, managing permissions, and even linking to decentralized finance (DeFi) protocols, allowing it to perform a variety of financial operations. This programmability empowers the AI to act as a fully autonomous agent, capable of managing assets without manual intervention, a capability rarely available in traditional finance.

“Autonomous agents can go off-rails, and with access to funds, the stakes are even higher,” Levin pointed out. “Keeping these agents aligned with their intended goals is a huge challenge. You need monitoring systems, budget constraints, and policies to prevent agents from delivering chaotic results that could jeopardize significant funds. The unpredictability of agents poses unique challenges, and without the right infrastructure, the risks grow exponentially with each transaction. That’s why robust monitoring and control mechanisms are crucial in this space.”

Furthermore, blockchain transparency and immutability offer a unique advantage. Every transaction executed by the AI is recorded on-chain, creating an auditable trail of activity that provides transparency and accountability—features highly valued by both investors and regulators. This makes crypto wallets a suitable infrastructure for autonomous agents in the finance world, provided that certain security and control measures are in place.

The Key Ownership Dilemma for AI Agents: Who Controls the Wallet?

Despite their advantages, crypto wallets introduce a critical challenge for AI agents: key ownership. For an AI agent to manage a crypto wallet, it needs access to the private keys. However, once an agent has these keys, it effectively has unrestricted control over the wallet’s assets, which poses several risks:

Unsupervised Spending: If an agent gains autonomous access, it could execute transactions that breach regulations or investment guidelines, potentially harming the interests of stakeholders.
Hacking and Security Risks: Storing private keys directly in the AI’s system makes them vulnerable to cyberattacks, especially if the AI operates in a networked environment.
Regulatory Compliance: Without a clear framework for oversight, regulators may hesitate to allow AI-managed wallets to operate, especially if there’s no mechanism for limiting unauthorized transactions.

“To reduce risk, we aggregate outputs from multiple AI models in a voting system, using diverse AI models and validators,” said Albert Castellana Lluís, co-founder and CEO of YeagerAI. “This helps us decentralize intelligence, minimizing errors and malicious manipulation by relying on collective decision-making. Essentially, it’s like putting the AI’s decisions to a jury vote.”

To address these challenges, several approaches to key management for AI agents have emerged, each with its own strengths and trade-offs.

Exploring Key Management Solutions for AI Agents Wallets

Multi-Signature Wallets: Multi-signature (multi-sig) wallets require multiple private keys to authorize a transaction, which can be useful for AI agents. For example, an AI agent could have access to one key, while human stakeholders or an external security protocol hold the others. Transactions would need multi-party approval, providing a layer of oversight and limiting the agent’s autonomy for high-stakes transactions.

Pros: Multi-sig is straightforward, well-supported in the crypto space, and provides transparency.
Cons: It may slow down transaction execution, and coordinating multiple signatures can be challenging in complex workflows.

The adoption of multi-sig wallets has seen significant growth, particularly with platforms like Safe. Initially designed as a multi-sig wallet, Safe has evolved into a comprehensive smart contract wallet, offering enhanced security and flexibility. This transition allows for more complex transaction logic and integration with decentralized applications, making it a robust solution for managing crypto assets.

Trusted Execution Environments (TEEs): TEEs are secure zones in hardware, such as within CPUs, that allow the AI to process sensitive data like private keys securely. By isolating the key within the hardware, TEEs prevent unauthorized access, even if the AI’s software environment is compromised.

Pros: TEEs offer high security by physically isolating the key.
Cons: Hardware limitations make scalability challenging, and not all crypto wallets support TEE integration.

Multi-Party Computation (MPC): MPC, a cryptographic approach where the private key is split into multiple “shares” held by separate parties, is emerging as a promising solution for autonomous agents. In this model, the AI only has partial control over the wallet key. When a transaction is proposed, the other parties’ approvals are needed to reconstruct the key and authorize the transaction. MPC offers a decentralized yet secure way to maintain control without restricting the agent’s independence entirely.

Pros: MPC provides robust control and a strong compliance framework, as each action requires multi-party consent.
Cons: MPC implementations can be costly and computationally demanding, though ongoing developments are improving efficiency.

Leading companies like Fireblocks have driven significant advancements in MPC infrastructure. Their platforms offer tools specifically designed for secure key management at an institutional scale, providing the speed and scalability needed for high-frequency transactions. This approach makes MPC increasingly viable for AI-based finance, enhancing both security and regulatory compliance.

Reflecting on his experience, Castellana Lluís added, “Our experience with StakeHound taught us that even the best cryptographic solutions fall apart without rigorous protocols. Avoiding human error is essential, and we’ve integrated that lesson into our current key management strategies. Stakeholder involvement is absolutely essential in designing secure key management strategies. Key management systems are only as effective as the processes we put in place to avoid human errors and ensure compliance.”

Multiple keys.

getty
Looking Ahead: Balancing Autonomy and Control for AI Agents

As AI agents gain ground in the world of finance, finding the right balance between autonomy and control is critical. Crypto wallets provide a viable foundation for AI-based finance, but effective key management remains a challenge. Solutions like multi-signature wallets, TEEs, and MPC each offer distinct advantages, and a layered approach using multiple solutions may ultimately prove most effective.

While these approaches make AI agents more viable in finance, regulatory questions remain. Agencies will need assurances of accountability and transparency, and the crypto industry will need to provide frameworks that protect against both security risks and misuse. For those interested in pioneering this space, exploring hybrid strategies and collaborating with regulatory bodies will be essential to bring autonomous AI agents to maturity.