Legal Quandary of Digital Avatars Deep Dive: When AI Attends Meetings, Signs Contracts, and Socializes for You—Who Bears Liability

Full Article

In May 2028, the European Parliament's Committee on Legal Affairs began reviewing the draft Digital Agent Act. This legislation aims to establish a legal framework for the increasingly prevalent AI digital avatar technology, representing the world's first legislative attempt specifically addressing digital avatar conduct.

AI digital avatars are AI agents trained on an individual's voice, appearance, behavioral patterns, and knowledge base, capable of attending meetings, responding to emails, conducting preliminary business negotiations, and even signing low-risk contracts on behalf of their human principal. Currently, products including Microsoft's Copilot Avatar, Google's Project Starline, and China's iFlytek "Digital Employee" have collectively served over 50 million users.

As user bases grow, legal disputes involving digital avatars are rapidly increasing. According to the EU legal database EUR-Lex, as of April 2028, over 200 legal cases worldwide involve digital avatar conduct. These cases concentrate in three areas: validity of contracts signed by digital avatars, defamation liability for statements made by digital avatars, and attribution of personal information leaked by digital avatars.

The most closely watched case occurred in March 2028: a French executive's digital avatar, while auto-replying to a business email, inadvertently confirmed a €3.2 million purchase order. The executive subsequently claimed ignorance and disagreement with the transaction, but the counterparty demanded contract performance. The case remains under review at the Paris Commercial Court.

The EU Digital Agent Act draft's core provisions include: First, a digital avatar's actions are legally attributed to the principal unless the principal can prove the action exceeded their pre-set authorization scope. Second, digital avatar developers and deployers bear joint liability for technical defects. Third, actions involving personal rights—marriage, wills, criminal confessions—may not be performed by digital avatars.

EDRi policy director Diego Naranjo cautiously welcomed the legislation but noted ambiguity in the "authorization scope" provision. "How do you define whether a digital avatar's action falls within authorized scope? If my avatar uses a tone I wouldn't use in an email, does that count as exceeding authorization? The law needs clearer standards."

In China, the NPC Standing Committee has included digital avatar legislation in its 2028 legislative work plan. Zhang Xinbao, professor at CUPL's Data Rule of Law Institute, said China's legislative approach may focus more on protecting the principal's rights. "In China's cultural context, people may be more concerned about their digital avatar 'saying something it shouldn't have.' Legislation needs to provide sufficient post-hoc remedies for the principal."

iFlytek VP Liu Cong raised concerns from an industry perspective: "If laws impose too many restrictions on digital avatars, it could stifle this emerging market. We suggest a tiered management approach—granting more autonomy for low-risk scenarios like internal meeting notes, while requiring strict human confirmation for high-risk scenarios like contract signing."

The establishment of a digital avatar legal framework is fundamentally answering an age-old legal question: when the agent is not a person but a machine, how should the rules of agency be adjusted? The EU's Digital Agent Act is expected to complete review by end of 2028, becoming the world's first enacted digital avatar-specific legislation.