生成式人工智能犯罪归责的开发者中心主义

Developer-Centric Liability of Generative AI Crimes

刑法介入智能犯罪的治理因应风险防范和安全控制的需求。生成式人工智能主体资格应予否定,人工智能系统本身无法成为刑事责任主体。不论未来人工智能技术发展程度如何,开发者作为风险源的创造者应当为风险现实化的危害结果担责,也最有能力控制危险源以避免犯罪。开发者责任始终是刑法规制人工智能风险的核心,通过建构以开发者为中心的刑事责任体系,有力规范生成式人工智能的行为。生成式人工智能的开发是指包括产品化研发及其投入应用的全过程,为人工智能系统创造了运行条件并引入市场的人类主体统称为“开发者”。由于开发生成式人工智能具有促进社会进步的技术创新价值,刑事归责应当平衡好安全保障和促进创新的需求,避免归责的盲目和刑罚的越界。容许风险理论则能兼顾二者,通过将开发者的技术创新区分为刑法不容许的风险行为和刑法容许的风险行为两类存在实质性差异的基础归责形态,以公正的风险分配理念指引开发者归责。开发者违背基本伦理准则或违反刑法规范而开发智能系统的,属于不容许的风险范畴,为刑法禁止并依据有关条款追究刑事责任。开发活动在风险容许范围内,但智能产品仍可能存在剩余风险,开发者可能对危害后果承担刑事责任,通常构成产品刑事责任或监督管理过失责任。

Criminal law’s involvement in intelligent crime governance is driven by the need for risk prevention and security control.Generative artificial intelligence(GAI)should not be considered a criminally liable subject,as intelligent systems cannot bear criminal responsibility.Regardless of AI’s future advancements,developers,as the creators of risk sources,should be accountable for the harmful consequences of realized risks and are best positioned to control these risks to prevent crimes.Developer liability is central to criminal law’s regulation of AI risks.Establishing a developer-centric criminal liability system effectively regulates generative AI behavior.Generative AI development encompasses the entire process from product research and development to application deployment.The human agents who enable AI systems to operate and introduce them to the market are termed“developers”.Since generative AI development fosters technological innovation and societal progress,criminal liability must balance security assurance with the freedom of innovation and avoid arbitrary liability and excessive punishment.The theory of permissible risk reconciles these needs by differentiating between impermissible and permissible risk behaviors under criminal law.This approach guides developer liability based on the principle of fair risk distribution.Development that violates basic ethical standards or criminal law norms falls within the category of impermissible risk and is prohibited and sanctioned by criminal law.The development activities are within an acceptable risk range.However,there may still be residual risks associated with smart products.Developers may bear criminal liability for harmful consequences,which typically constitutes criminal liability for the product or liability for negligence in regulatory oversight.