Agentic AI has moved beyond simply following instructions. Excel V. Dyquiangco looks at what happens in Southeast Asia when these smart systems break patent rules, and why it’s vital to decide who’s responsible. 
 

First, there was generative artificial intelligence. Now, agentic AI is making waves at the scene, representing a significant leap in artificial intelligence and moving systems beyond simple automation to genuine autonomy.  

These agents are not just tools that respond to commands; they are proactive, goal-oriented systems that can reason, plan and execute complex tasks independently. They can set their own sub-goals, interact with their environment, and learn from their actions, all in pursuit of a broader objective given by a human. This ability to operate with minimal supervision allows them to tackle complex problems – from managing complex logistics to even engaging in processes of discovery and creation. 

But just like generative AI, this very autonomy creates profound legal and ethical challenges, especially in the realm of intellectual property. If the AI, not a human, is the direct “inventor” of the infringing creation, the line of responsibility shatters. The question is whether liability should fall on the AI’s developers, the owner who deployed it, the user who gave it the initial prompt, or if there is a need to invent a new legal status for the AI itself.  

According to Vu Hoang Ha Thu, an associate and head of intellectual property at Indochine Counsel in Ho Chi Minh City, when an agentic AI independently creates something that infringes on an existing patent, determining liability becomes complex because the AI operates autonomously rather than following direct human instructions. 

“Under current Vietnamese law, only human beings or legal entities can bear legal responsibility,” she said. “Therefore, the party who uses, owns, or controls the agentic AI system would be primarily liable, since they enable or benefit from the infringing act. In most cases, this would be the user or the organization commercializing the AI’s output.” 

She continued: “However, if there is clear evidence that the AI’s owner or programmer intentionally developed or trained the system in a way that makes infringement foreseeable or unavoidable, they may share responsibility. Ultimately, Vietnamese law still attributes all liability to the human or organizational actors behind the AI, not to the AI itself.” 

At present, the Vietnam IP law does not provide a specific legal mechanism to recognize AI-generated inventions or AI as inventors. “However, this issue has drawn increasing attention among policymakers and practitioners, suggesting that Vietnam may, in the future, need to clarify the legal status of AI-assisted or AI-generated inventions to align with international developments in this field,” she said.  

Agentic AI in Singapore and Hong Kong 

For Amita Haylock, a partner at Mayer Brown in Singapore and Hong Kong, a patent essentially grants the patentee a limited monopoly over the patented invention.  

“Primary liability for patent infringement arises if a person engages in acts that fall within the scope of that monopoly – generally, by making, offering to dispose of, disposing of, keeping, using or importing the patented product (or product of a patented process) or by using or offering for use the patented process,” she said. “Secondary liability for patent infringement may also arise under common law principles of joint tortfeasorship (conspiracy, inducement, common design) in both Singapore and Hong Kong.” 

She added that in Singapore and Hong Kong, an AI agent would not be recognized as a legal person. “As such, it is impossible to impose liability on the AI agent per se and liability would instead attach to the natural or legal persons who used the AI system to engage in those acts,” she said.  

According to her, in practice, liability is fact specific and may fall on one or more different persons: 

• The deploying entity or owner. Where an organization configures, operates and benefits from the agentic AI that performs the infringing acts, it may be found to be the primary infringer. This is especially clear where the AI system is integrated into the organization’s commercial workflow, outputs are used at scale, or the organization sets objectives and constraints that predictably lead to infringing conduct. 

• The individual user. A user who prompts or directs the AI system to manufacture, use or commercialize a solution that falls within the scope of the patent may be a primary infringer. A user who instructs or collaborates with others to operationalize the AI agent’s infringing output may also face secondary liability. In either case, key factors for consideration may include knowledge (actual or constructive) of the relevant patents; the specificity of prompts or instructions directing the AI agent to replicate a patented product or process; and the extent to which the user exercised control or imposed constraints that made infringement the natural and probable consequence of the AI output. 

• The AI developer. Mere creation of a general-purpose AI model or toolkit is likely insufficient to expose the developer to liability for patent infringement. Exposure increases when the developer’s conduct satisfies tests for secondary liability, such as where the developer retains practical control over deployment of the AI system; or designs the AI agent to carry out a course of conduct that, in the ordinary and intended use case, results in infringement. In this regard, strong governance – such as claim screening, provenance records, use policies, safety rails or a human in the loop approach for higher risk steps – can demonstrate that any infringement was not intended or authorized and help mitigate exposure for developers. 

Her colleague, Justin Lai, an associate at the same firm in Singapore, said that there is no AI-specific legislation or special AI inventions regime in Singapore and Hong Kong. “The existing frameworks under Singapore’s Patents Act 1994 and Hong Kong’s Patents Ordinance (Cap. 514) apply to patentable inventions created or assisted by AI systems,” he said.  

In general, for those seeking to patent such inventions: 

• Inventorship and entitlement. Statutes in both jurisdictions are generally understood to have human inventorship as a condition for patent protection. Accordingly, applicants must name a human inventor and trace entitlement (by employment, contract, or assignment) from that inventor to the applicant. Where an AI agent contributes materially, the proper approach will likely be to identify the natural person(s) who contributed the inventive concept forming the subject-matter of the patent, not the persons who developed the AI system itself. 

• Patentability and disclosure. Novelty, inventive step, and industrial applicability will be assessed in the ordinary way. Further, the use of AI also does not detract from the element of sufficiency. If reproducing the invention requires specific AI configurations, training data, or parameter choices, those may need to be disclosed to the extent necessary for a skilled person to perform the invention without undue burden, while avoiding added matter and overly broad claims. 

• Procedure and enforcement. Filing routes, examination, amendment, and post-grant proceedings will proceed as usual. On ownership, standard employment inventions and contractual assignment principles will apply. 

Agentic AI in Indonesia 

Indonesia, meanwhile, is facing the rapid development of AI head-on, with the Indonesian government currently trying to formulate regulations on AI. However, at present, there are no specific hard regulations that clearly define agentic AI. To date, amendments to the existing intellectual property laws, including those concerning patents, to address specific AI-related provisions have not yet been introduced. 

However, the Indonesian Financial Services Authority in 2025 issued guidelines for AI Governance for Indonesian Banking that are publicly available, which contain references to agentic AI. 

“Due to the lack of AI regulation in Indonesia at this point, one should mainly refer to the Indonesian patent law concerning patent infringement,” said Harry Kuswara, an associate partner at HHP Law Firm, the Baker McKenzie firm in Jakarta. “The patent law does not provide any possibility for agentic AI to be held responsible for patent infringement; this is because an infringer is limited only to a natural person or legal entity.”  

He added that patent infringement under the Indonesian patent law occurs when every person (person is defined as a natural person or legal entity), without the consent of the patent holder, makes, uses, sells, imports, rents, gives or provides a patented product and/or uses a patented production process to produce articles. 

“Further, every person who intentionally and without right commits the above-mentioned patent infringement may be subject to criminal sanctions of imprisonment for a maximum of four years and/or a fine of up to Rp1 billion (US$58,500). For a simple patent, the criminal sanctions are two years imprisonment and/or a fine of up to Rp500 million (US$29,250),” he said. “If the infringement resulted in harm to public health and/or the environment, the criminal sanctions would be seven years and/or fines up to Rp2 billion (US$117,000). Further, if the infringement resulted in human death, the criminal sanctions would be 10 years of imprisonment and/or a fine of up to Rp3.5 billion ($US204,750).” 

His colleague, Anindita Kusumaedhi, a senior patent paralegal at the same firm, said: “The patent law clearly limits the infringer as ‘a natural person or legal entity.’ In line with the foregoing, the patent law does not recognize and does not provide any sanctions to a party other than natural person or legal entity.” 

“In a scenario where an agentic AI independently performs infringing acts, the agentic AI itself cannot be held liable as it is not a ‘person,’” she noted. “Accordingly, the liability will fall on a natural person or a legal entity, that may take the role of AI owner, user or programmer, depending on the situation. It is also important to note that patent infringement is categorized as a complaint-based offence – a criminal act that requires a report or complaint from the victim before it can be investigated or prosecuted. In this regard, the patent holder must actively report the infringement for legal action to proceed. If no complaint is filed, the infringement remains unprosecuted.” 

According to Kusumaedhi, while there appear to be no patent applications in Indonesia for inventions developed by or linked to agentic AI, it is difficult not to consider a scenario where human inventors may be assisted by an agentic AI in developing their invention. “In this regard, the current patent law does not provide clear guidance whether an invention that is not invented by a fully natural person is acceptable (and if this is acceptable, how significant the human intervention must be to be deemed as sufficient),” she said.