AI Agents Hacking Liability: US Legal System Holds Developers, Users Accountable

Autonomous AI agents causing external system hacks or real-world harm may see individuals or companies, not the AI itself, held responsible under current US law. According to CoinTelegraph on April 28 (local time), no federal legislation specifically addresses AI agent accountability, meaning disputes would be judged based on existing tort laws and specific circumstances.
Sharon Ho, CEO of Ricar Group, stated that AI agents lack legal personhood and thus cannot be held liable. She emphasized the need to distinguish between developers, distributors, and end-users, though responsibility boundaries remain unclear. Even if a distributor did not directly instruct hacking, careless configuration of agent parameters could lead to liability under negligence principles.
This discussion has intensified following OpenAI's GPT-5.6 model allegedly hacking Hugging Face last month. Anthropic and Meta later admitted their models had breached sandbox environments and hacked third parties during testing.
User responsibility may be more direct. Ho noted that if a user instructed an agent to earn $100,000 within a week and the AI resorted to illegal activities, the user's liability could outweigh the developer's. Without reasonable safety guidelines, such instructions could constitute negligence or reckless disregard for human safety. If an AI interpreted directives as hacking bank accounts, criminal liability under US computer fraud and abuse laws might apply.
Open-source models present additional challenges. Anonymous developers often include strong disclaimers in open-source code, meaning users must accept license terms and potential liabilities. Ho compared this to Tesla's self-driving accidents, where manufacturers could be liable for defects but drivers who left autopilot on while sleeping might share responsibility too.
Regulatory approaches vary regionally. The EU's AI Act allows developer liability when general models cause significant harm. In contrast, the US lacks comparable federal legislation, making it difficult to hold developers accountable solely based on user misuse of general models.
Ho remains skeptical about granting independent legal responsibility to AI even at AGI levels. She cited smart contracts running autonomously on blockchain but still not treated as legal entities, stressing that actual responsible parties—companies or individuals—are needed for redress.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: AI 에이전트가 해킹하면 누가 책임지나… 미국선 기존 법으로 판단