According to Woofun AI, the Bedrock AgentCore Payments platform jointly developed by Coinbase (COIN.US) and Stripe has enabled AI agents to complete identity verification and payments within preset limits using stablecoins and the x402 protocol.

Although these wallet-enabled shopping agents have already been placing orders autonomously across the open internet, and merchants have not yet faced large-scale legal action, the question of who bears responsibility for refunds after successful payments is rapidly becoming a core industry conflict. Edgars Nemse, CEO of the GenLayer Foundation, told CryptoSlate that while the payment process is deterministic, the subjective nature of service delivery makes dispute resolution extremely complex, which directly limits the purchasing capability of agents. All current dispute resolution systems are built on the implicit assumption that "cumbersome complaints cause users to give up," and AI is removing this barrier, leading to a surge in complaint volumes that traditional human processing teams are approaching their limits in handling.

As AI eliminates the threshold for filing complaints, global chargeback costs are facing exponential growth risk. The Reserve Bank of Australia documented this trend in its October 6 summary of conclusions on its payments system consultation, citing input from 75 stakeholders. Data shows that complaints filed with the U.S. Consumer Financial Protection Bureau doubled to 6.6 million in 2025, and the agency warned that large language models (LLMs) and autonomous software will cause a flood of duplicate complaints. Research published in the journal Nature Human Behaviour found that using LLMs can increase the probability of obtaining a favorable ruling by 6.9 percentage points. Data compiled by Woofun AI shows that Mastercard (MA.US) and Datos predict global chargeback transactions will reach 324 million by 2028. Based on a baseline cost of $128 per transaction for U.S. merchants in 2026, if global costs rise 5%, it would generate 16.2 million additional chargebacks and $2.1 billion in operational costs; if costs rise 15%, additional chargebacks would reach 48.6 million, with costs as high as $6.2 billion. Merchants, payment service providers, and card issuers all say that existing rules cannot clearly determine liability when agents exceed their authority, and that AI-assisted purchases require an additional 4% fee. All parties favor establishing industry standards before the Reserve Bank of Australia publishes its regulatory priorities by the end of 2026.

The battle for interface control between platforms is intensifying merchants' difficulties, as Amazon.com (AMZN.US) banned Meta Platforms, Inc. (META.US)'s Muse shopping agent on grounds of unauthorized access. Nemse analyzed that this move is not a technical limitation, but rather Amazon.com (AMZN.US) seeking to retain control over billions of dollars in customer relationships, data, and advertising resources, preventing its interface from becoming a generic API. Alphabet (GOOGL.US) faces the same dilemma and is expected to block external agents to promote its own products. In contrast, smaller merchants are more inclined to find partners who can solve their problems, and Shopify has already allowed browser-based AI agents to enter the checkout process. However, platforms control the interfaces and adjudication power, meaning small merchants still need to turn to Amazon.com (AMZN.US) even after identifying problems. A CI&T survey of 1,011 U.S. consumers showed that 27% are willing to accept a fully AI-driven shopping experience, but the limit of the agent-driven model lies in the degree of unremedied loss that consumers can tolerate. Since API calls cost only a few cents, agents need to pay for those calls, but in the absence of clear accountability pathways, no one allows them to make unilateral decisions on refunds or insurance claims. Solutions such as Alphabet (GOOGL.US) AP2, Mastercard (MA.US) Agent Pay, and Visa (V.US) Intelligent Commerce focus on authorization management, including signed instructions, tokenized credentials, and identity verification, but they can only prove what the buyer requested and cannot determine delivery outcomes.

To address these pain points, GenLayer has proposed an on-chain consensus-based adjudication mechanism aimed at resolving subjective disputes through technical means. The solution has validators run large language models, determine final outcomes through a consensus mechanism, execute them on-chain, and allow relevant parties to appeal. GenLayer says common disputes can be resolved within 30 minutes, while complex cases reach conclusions within 3 hours. Since AI agents can initiate disputes at zero cost, the system must introduce fees, deposits, or reputation penalties to make frivolous disputes uneconomical. Validators will make judgments based on evidence such as receipts, logistics tracking, and task requirements. The appeal mechanism adds time and cost but can avoid poor model outputs. On-chain adjudication results will be used to manage escrowed funds, but ordinary merchants' credit card chargebacks are not covered. The core of this mechanism lies in establishing a reliable and neutral dispute resolution system to fill the gap left by platform adjudication power, ensuring that after agents find small merchants, those merchants no longer need to rely on the protection of large platforms, thereby achieving fair transactions at the technical level. The establishment of industry standards will determine the ultimate direction of AI business models.

If merchants and platforms can reach agreement on standards, establishing verifiable authorization instructions, a comprehensive evidence recording system, and escrow mechanisms that can intervene before disputes escalate into chargebacks, AI agents will expand from simple API calls to various service interactions with unfamiliar counterparties. This will give small merchants the same right to hold parties accountable as large platforms, breaking platform monopolies. Conversely, if the cost of filing disputes is low while the cost of resolving them is high, merchants will be forced to raise fees, restrict agent purchasing behavior, or push buyers back to trusted traditional platforms. Nemse emphasized that a better payment system cannot change this limitation; the key lies in clear accountability pathways and responsible parties. As regulatory priorities approach at the end of 2026, all parties need to find a balance between interface control and an open ecosystem. Otherwise, the AI agent-driven business model will stall due to a lack of trust, and massive chargeback costs will become a sword of Damocles hanging over the industry.