Anduin Framework Enhances Sybil Resistance in AI Agent Marketplaces

AI agents are increasingly engaging in autonomous transactions, leveraging payment standards like x402 to access and pay for services. However, current reputation systems in agentic marketplaces are highly vulnerable to Sybil attacks, where adversaries manipulate reputation scores using fake identities. A recent study introduces the Anduin framework, which uses payment graph analysis to enhance trust and mitigate these vulnerabilities.

Challenges in Existing Reputation Systems

Current reputation mechanisms for AI agents fall into four main categories, each with significant limitations:

• Curated Marketplaces and Allow-Lists: These provide strong assurances but have limited reach due to the time-intensive review process.
• Per-Transaction Evaluators: These involve third-party verification but incur high costs, making them unsuitable for sub-cent micropayments.
• Voluntary Onchain Ratings: Data sparsity is a major issue, with fewer than 4,000 ratings recorded across 66 million x402 payments.
• Raw Usage and Volume Metrics: These are easily manipulated through self-directed payments, costing attackers as little as $0.017 per transaction.

An empirical study found that elevating a fake provider's reputation to the marketplace median costs just $0.0027, approximately 1/259th of what an honest median provider earns. This highlights the need for more robust systems.

Anduin Framework: A Payment-Graph-Based Solution

The Anduin framework leverages public blockchain records to derive reputation from payment graphs. Every x402 payment creates a cryptographically signed, permanent record of value transfer. By mapping these transactions, Anduin constructs a bipartite graph connecting buyers and providers, weighted by monetary value, transaction frequency, and recency.

Trust is localized and relational, based on overlapping commercial activity between buyers and providers. For example, if a buyer evaluates a new provider, the framework analyzes payment paths involving co-users who share trusted providers. This approach ensures that trust signals are grounded in real economic activity, making it significantly harder for attackers to fake reputation.

Cost of Manipulation and Sybil Resistance

Under the Anduin framework, Sybil attacks become prohibitively expensive. Fake accounts must spend real money at legitimate providers trusted by the target buyer, as self-directed payments between wallets controlled by the attacker do not contribute to reputation. The study found that manipulating Anduin's reputation system is 23 times more expensive than current methods, making it a robust solution against Sybil attacks.

Implications for Agentic Marketplaces

By integrating Anduin, marketplaces can enhance trust without relying on centralized curation or voluntary feedback. This decentralized approach aligns with the principles of blockchain technology, offering a scalable and cost-effective solution for reputation management in rapidly growing agentic economies.