IMDA discussion paper explores legal responsibility for AI agents

What makes agentic AI different

The Paper identifies three features of agentic AI systems that are particularly relevant to liability:

1. Autonomy. Agentic AI systems operate with reduced human intervention between instruction and outcome. The less a human is in the loop, the harder it is to attribute the AI agent’s actions and their consequences to a specific person or organisation.
2. Decision-making. Unlike traditional generative AI, which produces outputs for a human to act on, AI agents decompose tasks, choose between alternatives, and adapt when initial approaches fail. This increases the scope for unexpected or misaligned behaviour.
3. Action-taking and tool use. AI agents interact with external systems: executing database queries, browsing the web, making API calls, and, increasingly, operating computers in ways similar to humans. The range of tools determines the AI agent’s “action space”, and with it, the potential impact when something goes wrong.

The Paper flags that as agentic AI systems are becoming more independent and wide-ranging, the liability rules should reflect where the technology is going, not just where it is now.

Too many actors in agentic AI systems

A core challenge identified in the Paper is the proliferation of actors in the agentic AI value chain. A single agentic AI system may involve:

This creates two interrelated problems. First, there’s a question of principle: even with all the facts, it may still be unclear who is responsible, or by how much. Second, there’s a practical difficulty: in many cases, it may be impossible to establish the full picture, given time cost limitations, and commercial secrecy. 

How existing law applies 

A majority of the Working Group considered that existing common law may address many liability scenarios arising from agentic AI systems; but acknowledged significant practical challenges. The Paper further examines several legal principles:

Contract

Contracts are an effective means of pre-allocating risk between actors in the value chain. However, their usefulness is limited by the doctrine of privity: third parties harmed by an agent generally cannot enforce contractual protections agreed between others. 

More fundamentally, parties with weaker bargaining power, particularly consumers, risk having most of the liability disclaimed and pushed onto them through onerous terms of use. The Paper observes that without intervention, every actor in the chain will rely on disclaimers, with the burden ultimately falling on end users.

Tort of negligence

Negligence claims face several difficulties in the agentic AI context. First, establishing a duty of care may be problematic where the relationship between the actor and the claimant is too remote – e.g., a system provider may not owe a duty to every third party whose data is affected by an AI agent’s actions. 

Second, considering breach and determining the “reasonable” level of safeguards is difficult when AI agents are designed to operate autonomously and their behaviour evolves beyond the developer’s specifications. 

Lastly, causation is complicated by the number of actors involved and the opacity of non-deterministic systems. Compounding the issue of remoteness, this raises the question of whether an AI agent’s unexpected behaviour was reasonably foreseeable at all.

Product liability

While product liability could help apportion responsibility in favour of end users, Singapore’s product liability laws are currently limited to narrowly defined consumer goods and do not cover losses arising from AI. 

The Paper notes that the EU’s expanded Product Liability Directive, which will cover AI systems placed on the market after 9 December 2026, provides an interesting comparison.

Areas for further study

Looking forward, the Paper identifies three areas for further study:

1. How should responsibilities be allocated along the value chain? 

Model developers have the greatest control over base behaviour but limited visibility into deployment context. Deployers and end users know the use case but cannot alter the model’s core tendencies. 

The Paper suggests a spectrum: developers expected to mitigate general or baseline risks, and deployers responsible for use-case-specific safeguards, with standardised disclosures as a complementary mechanism to make this workable.

2. How can actors with limited bargaining power be better protected? 

Left to the free market, weaker parties, particularly consumers, will absorb disproportionate risk. 

The Paper calls for further study on measures such as simplified and expedited dispute resolution mechanism, legal or evidential presumptions (e.g., shifting the burden of proof to make it easier for claimants to obtain proof, considering information asymmetries), and sector-specific liability frameworks.

3. Who bears responsibility for truly unforeseeable AI agent actions? 

Where all actors have taken reasonable safeguards, but the AI agent nevertheless behaves unpredictably and causes harm, the loss currently falls where it lands. 

The Paper suggests that factors such as transparency, the distribution of benefits across the value chain, and the reasonableness of reliance placed on the AI agent should inform any future allocation.

Practical takeaways

The Paper avoids setting out specific policy fixes, but its analysis points to some clear practical implications:

For developers and system providers: 

• Focus on meaningful safety testing, put effective guardrails in place, and be open about the system’s limits;
• Broad disclaimers that try to push all risks onto users are likely to face scrutiny and may not hold up if challenged;
• Clear disclosures about known risks, gaps in instruction-following, and foreseeable failures will matter more in showing that reasonable care was taken.

For deployers: 

• The use case is critical. Putting an AI agent into a high-risk setting such as financial services, healthcare, or consumer-facing roles without safeguards beyond the developer’s baseline will be difficult to justify. 
• Oversight should match the level of risk. Routine, low-risk tasks may need lighter monitoring, but more significant decisions should still involve human approval.

For enterprises using AI agents: 

• Contracts with AI providers should deal with unexpected behaviour, not just known risks. 
• It is also worth thinking carefully about how far liability exclusions will hold, particularly in business-to-consumer settings where consumer protection rules may limit them. Internally, it is important to have clear policies on oversight, escalation, and incident response.

For all actors: 

• Keep a close watch on developments. 
• Singapore is among the first jurisdictions to look at agentic AI system liability in a structured way, and parallel developments in other jurisdictions, such as the EU, are likely to influence the direction of travel. 
• Organisations that engage with these issues early will be better prepared than those that react later.

Conclusion

The Paper does not call for immediate regulation, but instead maps the legal challenges posed by agentic AI systems. Its key message is that while existing legal doctrines can address many scenarios, the combination of multiple actors, non-deterministic behaviour, and greater autonomy creates real gaps, especially for consumers and third parties.

For now, traditional legal principles remain the main tool for dealing with AI-related harm. However, as AI agents become increasingly sophisticated and widely deployed, questions around liability (its potential and scale), accountability, and access to remedies compound and will become harder to resolve and more important in practice.

The Paper signals the direction of future policy. Organisations using or developing AI agents should watch these developments closely, as they are likely to shape regulatory expectations, litigation risk, and governance standards in Singapore and beyond.

For further guidance please reach out to the authors of this alert or your usual Hogan Lovells contact.

Authored by Charmian Aw, Ciara O'Leary, and Florence Seow.