Privacy & Legal in Agentic Commerce
Agentic Commerce raises fundamental legal questions: Who is liable when an agent makes a wrong purchase? What does consent look like when no human is at the checkout? And what does GDPR mean for AI-powered transactions?
Note: This article provides an assessment of current legal questions and does not constitute legal advice. For specific legal decisions, please consult a qualified attorney.
Who Is Liable for Erroneous Purchases?
The central question: If an AI agent orders the wrong product, purchases a product that is too expensive, or triggers an order the user did not want — who bears the responsibility?
Under current law in Germany and the EU: The user. An AI agent has no legal personality of its own. It acts as an instrument on behalf of the user — comparable to an authorized third party. The contract is concluded between the user (buyer) and the merchant (seller).
The liability risk therefore lies with the user, underscoring the importance of consent models and confirmation mechanisms. Currently, all Agentic Commerce implementations require an explicit purchase confirmation before the transaction.
GDPR and Agentic Commerce
In Agentic Commerce, personal data is processed at multiple points:
- At the agent provider: Conversation histories, preferences, purchase history, payment information
- At the merchant: Shipping address, order data, payment processing
- At the payment provider: Tokenized payment information
Each of these actors is a separate controller under the GDPR and must document the legal basis, purpose and storage duration. For users, this means: you have rights of access, rectification and erasure against each actor.
Consent: Authorization for Agents
Consent in Agentic Commerce has two dimensions:
- Data protection consent (GDPR): Consent to the processing of personal data by the agent provider. Must be informed, voluntary and revocable.
- Transaction consent: Authorization of the agent to make a purchase. Currently via explicit confirmation (ACP) or through mandates with defined limits (AP2).
The GDPR requires clear, understandable consent. With mandates (AP2), it could be argued that the creation of the mandate itself constitutes informed consent — provided the conditions are transparently presented.
PSD2 and Strong Customer Authentication
The European Payment Services Directive PSD2 requires Strong Customer Authentication (SCA) for electronic payments: two of three factors (knowledge, possession, inherence) must be confirmed.
For agent transactions, the question arises: Who authenticates — the user or the agent? Possible approaches:
- Delegation: The user authenticates once when creating the mandate — every subsequent agent transaction is covered by the mandate
- Per-transaction SCA: For each purchase, the user is prompted for confirmation (e.g., via biometrics on a smartphone)
- Risk-based exemptions: PSD2 allows exemptions for low amounts and trusted recipients
Data Sovereignty: Who Owns the Customer Data?
A critical issue: When customers buy through an agent instead of the merchant's website, the merchant loses valuable data — browsing behavior, cart history, preferences.
The Trusted Agentic Commerce Protocol (TACP) by Forter addresses this problem: customer data is passed to the merchant via JWE encryption, enabling personalization and fraud detection — without compromising user privacy.
Consumer Protection
- Right of withdrawal: The 14-day right of withdrawal for distance contracts applies without restriction to agent purchases
- Information obligations: The merchant must inform about price, features and right of withdrawal before contract conclusion — the agent must display this information to the user
- Price transparency: The agent must display the final price including taxes and shipping before the user confirms
Regulatory Outlook
- EU AI Act: AI agents that make autonomous purchasing decisions could be classified as "High-Risk AI Systems", entailing transparency, documentation and oversight obligations
- Digital Markets Act: If major platforms (Google, OpenAI) are classified as "Gatekeepers", special interoperability and fairness obligations apply — including for agent commerce services
- Product Liability Directive: The revised EU Product Liability Directive could classify AI agents as "products", which would clarify liability questions
Frequently Asked Questions
Does the right of withdrawal apply to agent purchases?
Yes. The 14-day right of withdrawal for distance contracts applies regardless of whether the purchase was made manually or via an AI agent. The contract is concluded between the buyer and the merchant — the agent is merely an instrument.
Is the user or the agent responsible for erroneous purchases?
The user. An AI agent acts on behalf of the user but has no legal personality of its own. Erroneous purchases fall under the user's risk — hence the importance of consent models and confirmation mechanisms.
May an agent store my browsing data?
Only with your consent under GDPR Art. 6. Agent providers (OpenAI, Google) must transparently document which data is processed and for what purpose. You have the right to access and deletion.
What impact does the EU AI Act have on Agentic Commerce?
AI agents that make autonomous purchasing decisions could be classified as "High-Risk AI Systems". This would require transparency obligations, risk management and human oversight. The exact classification is still being discussed.