1. Introduction

AI-Based Conflict Prevention Infrastructure Liability in Singapore concerns the civil liability arising when artificial intelligence systems designed to identify, predict, prevent, or manage conflicts cause harm through negligence, inaccurate predictions, discriminatory decisions, system failures, or improper interventions.

Such infrastructure may be deployed by government agencies, private companies, transport operators, commercial property developers, employers, financial institutions, and public-service providers.

Examples include:

Predictive conflict-detection systems: AI identifies potential disputes, confrontations, or dangerous interactions before they occur.

Smart traffic-management infrastructure: AI coordinates traffic signals and detects situations likely to cause road accidents or disputes between road users.

Public-space monitoring: AI analyses crowd movements and identifies potential violence, overcrowding, or confrontations.

Automated dispute-prevention platforms: AI identifies contractual, commercial, or workplace conflicts and recommends preventive interventions.

Critical-infrastructure coordination: AI monitors electricity, water, transport, and telecommunications networks to prevent operational conflicts and cascading failures.

AI-assisted emergency management: Automated systems allocate emergency resources and coordinate responses to prevent escalation.

The central legal question is:

Who is legally responsible when infrastructure intended to prevent conflicts instead creates, aggravates, or fails to prevent foreseeable harm?

Singapore does not currently have a separate, comprehensive civil-liability regime specifically governing AI-based conflict-prevention infrastructure. Liability must therefore be assessed under established principles of negligence, contract, occupiers' liability, statutory obligations, data protection, and other applicable areas of Singapore law.

2. Legal framework governing AI infrastructure liability in Singapore

A. Negligence under Singapore tort law

Negligence is likely to be the principal basis of civil liability when an AI-based conflict-prevention system causes foreseeable harm.

A claimant generally needs to establish:

Duty of care: The defendant owed the claimant a legal duty to exercise reasonable care.

Breach of duty: The defendant failed to meet the applicable standard of reasonable care.

Causation: The breach caused the claimant's injury or loss.

Damage: The claimant suffered legally recognised and recoverable damage.

The leading authority is Spandeck Engineering (S) Pte Ltd v Defence Science & Technology Agency [2007] SGCA 37.

The Court of Appeal established a three-stage framework:

Factual foreseeability of harm.

Sufficient legal proximity between claimant and defendant.

Policy considerations that may negate or limit the duty of care.

This framework is particularly important where an AI infrastructure operator provides services to members of the public who have no direct contractual relationship with the technology provider.

For example, if an AI crowd-management system incorrectly identifies a peaceful gathering as a potential violent confrontation and directs security personnel to intervene, the injured person may seek to establish negligence against the operator, depending on the circumstances.

B. Contractual liability

Contractual liability may arise between:

AI infrastructure developers and government agencies.

Technology vendors and infrastructure operators.

Smart-city service providers and commercial property owners.

AI maintenance companies and transport operators.

Private security companies and their clients.

A contract may impose obligations concerning:

System reliability.

Accuracy of conflict predictions.

Maintenance and software updates.

Cybersecurity.

Human supervision.

Incident reporting.

Emergency response.

Allocation of responsibility for system failures.

Under Singapore contract law, a party may be liable for breach where it fails to perform an applicable contractual obligation.

Importantly, an AI supplier cannot automatically avoid contractual responsibility merely because the failure originated in a machine-learning model or an external software component.

C. Personal Data Protection Act 2012

AI-based conflict-prevention infrastructure may process:

Facial images.

Identification information.

Location information.

Behavioural patterns.

Video surveillance records.

Personal communications.

Information about interactions between individuals.

The Personal Data Protection Act 2012 (PDPA) is relevant where organisations collect, use, disclose, or otherwise process personal data.

Relevant obligations include:

Consent and applicable exceptions.

Purpose limitation.

Notification.

Accuracy of personal data where required for relevant decisions.

Protection of personal data.

Retention limitation.

Accountability.

An organisation that uses inaccurate personal information to classify someone as a potential threat may face regulatory consequences and, where the legal requirements are met, civil consequences.

However, a PDPA breach does not automatically establish a negligence claim or a general right to compensation for every form of AI-related harm.

D. Occupiers' liability and infrastructure safety

Where AI systems operate within shopping centres, transport facilities, residential buildings, workplaces, or other premises, the occupier's duty of reasonable care may be relevant.

For instance, an AI-powered crowd-control system might incorrectly close emergency exits, create dangerous congestion, or direct people towards an unsafe area.

The responsible occupier, technology provider, maintenance contractor, or security operator may face liability depending on their respective duties and degree of control.

E. Protection of constitutional and administrative interests

Where public authorities deploy AI to prevent social conflict, public disorder, or threats to public safety, questions of legality, fairness, procedural safeguards, and public accountability may arise.

The constitutional and administrative-law context is important, particularly where automated systems affect freedom of movement, public participation, access to public services, or the exercise of public powers.

However, constitutional or administrative unlawfulness and private-law damages are separate questions. A claimant must establish an applicable cause of action to obtain civil compensation.

3. At least seven important Singapore case laws

The following are genuine Singapore decisions. They do not concern AI-based conflict-prevention infrastructure directly; they provide established legal principles that can be applied by analogy to disputes involving such systems.

Case 1. Spandeck Engineering (S) Pte Ltd v Defence Science & Technology Agency [2007] SGCA 37

Leading authority

Facts and background:

The dispute concerned a construction contractor and the Defence Science & Technology Agency. The contractor alleged negligence in the performance of certification and supervisory responsibilities, despite the absence of a direct contractual relationship with the defendant.

Legal principles:

The Court of Appeal established the Singapore test for determining whether a duty of care exists in negligence claims.

The test examines:

Factual foreseeability.

Legal proximity.

Policy considerations.

Application to AI infrastructure:

Suppose a company develops an AI system for identifying potential confrontations in public transport stations.

The system repeatedly misclassifies ordinary passenger behaviour as threatening, resulting in unnecessary security interventions and physical injuries.

A claimant may argue that the developer or operator owed a duty of care if the relationship, reliance, control, and surrounding circumstances establish sufficient proximity.

Importance: This case provides the foundational test for deciding whether AI developers, system operators, and infrastructure managers owe duties of care to affected individuals.

Case 2. RSP Architects Planners & Engineers v Management Corporation Strata Title Plan No 1075 [1999] SGCA 30

Facts and background:

The dispute involved alleged negligence by architects in the design and supervision of condominium construction.

The management corporation pursued a negligence claim concerning defects in the building.

Legal principles:

The Court of Appeal considered the responsibility of construction professionals towards persons who may suffer loss because of negligent professional work.

The decision is important in understanding professional responsibility and liability for defective infrastructure.

Application to AI infrastructure:

An AI conflict-prevention system may be integrated into a building's:

Access-control systems.

Fire-safety infrastructure.

Emergency evacuation arrangements.

Surveillance networks.

Building-management platforms.

If negligent design or integration makes the infrastructure unsafe, the responsible professionals and companies may face negligence claims.

For example, an improperly designed automated access-control system might lock occupants inside a building during an emergency.

Importance: The decision supports the examination of professional responsibility for infrastructure design and supervision, including the design and integration of AI-enabled systems.

Case 3. See Toh Siew Kee v Ho Ah Lam Ferrocement (Pte) Ltd [2013] SGCA 29

Facts and background:

This case concerned a worker who suffered injuries during an industrial operation involving a crane and the movement of a vessel.

The Court of Appeal examined the responsibilities of occupiers and the relationship between control of premises and the activities taking place there.

Legal principles:

The judgment provides important guidance on occupiers' liability, negligence, and the significance of control over premises and activities.

It explains that an occupier's responsibility cannot be determined merely by identifying who owns or leases the property.

Application to AI infrastructure:

Consider a shopping centre using AI-powered crowd-monitoring technology.

The system detects a possible confrontation and automatically redirects visitors towards a restricted passage.

If the redirection creates a dangerous situation, the court may examine:

Who controlled the premises?

Who controlled the AI system?

Who supervised the security personnel?

Who had responsibility for maintaining safe access?

Whether the danger was reasonably foreseeable.

Importance: The decision is relevant to identifying the responsibilities of occupiers and operators of AI-enabled public infrastructure.

Case 4. Tan Juay Pah v Kimly Construction Pte Ltd [2012] SGCA 17

Facts and background:

The litigation arose from the collapse of a tower crane at a construction site at the National University of Singapore Society Kent Ridge Guildhouse.

The case involved questions concerning negligence, statutory responsibilities, and liability for construction-related risks.

Legal principles:

The Court of Appeal clarified that the existence of a statutory duty does not automatically create a corresponding common-law duty of care.

The existence of a common-law duty must be assessed using the Spandeck framework.

Application to AI infrastructure:

Suppose a government agency requires an AI infrastructure provider to comply with technical safety requirements.

If the provider violates a regulatory requirement and an individual suffers harm, the claimant must still establish the appropriate legal basis for civil liability.

Regulatory non-compliance may be relevant evidence, but it does not automatically establish every element of negligence.

Importance: This case is significant for distinguishing regulatory compliance obligations from private civil liability in AI infrastructure projects.

Case 5. Animal Concerns Research & Education Society v Tan Boon Kwee [2011] SGCA 2

Facts and background:

The dispute concerned the responsibilities of a clerk of works in relation to construction supervision and alleged negligence.

The Court of Appeal considered the circumstances in which a professional may owe a duty of care to someone with whom there is no direct contractual relationship.

Legal principles:

The judgment applied the Spandeck approach and examined proximity, assumption of responsibility, and reliance.

It demonstrates that professional responsibilities may extend beyond the immediate contractual relationship in appropriate circumstances.

Application to AI infrastructure:

An AI system integrator may undertake to monitor the reliability and safety of a conflict-prevention platform.

If the integrator assumes responsibility for identifying dangerous system failures but negligently ignores repeated warnings, an affected claimant may attempt to establish a duty of care.

Importance: The case is relevant to determining whether AI consultants, system integrators, and monitoring professionals owe responsibilities to third parties.

Case 6. Singapore Shooting Association v Singapore Rifle Association [2017] SGHC 266

Facts and background:

The case concerned two flooding incidents at the National Shooting Centre.

The dispute involved drainage infrastructure, maintenance responsibilities, structural problems, and the responsibilities of the occupier.

The High Court found that the first flood was not proved to have been caused by the defendant's breach, but held the defendant liable for losses of $4,708 resulting from the second flood.

Legal principles:

The judgment examined:

Infrastructure maintenance.

Occupiers' duties.

Control over premises.

Reasonable precautions.

Expert engineering evidence.

Causation.

Foreseeability of recurring harm.

Application to AI infrastructure:

This decision has particular relevance to AI-based conflict-prevention infrastructure.

Imagine that an AI system monitors a city's drainage, electricity, or transport networks to prevent operational disruptions that may escalate into public safety incidents.

If repeated warnings about system failures are ignored and an infrastructure breakdown occurs, the operator's failure to investigate, maintain, or take reasonable precautions may become relevant to a negligence claim.

The case also illustrates that a claimant must prove that the defendant's breach actually caused the loss.

Importance: This is a particularly useful infrastructure-liability analogy because it deals with real physical infrastructure, maintenance failures, and recurring risks.

Case 7. Ngiam Kong Seng and Another v Lim Chiew Hock [2008] SGCA 23

Facts and background:

The litigation concerned a road accident and claims involving psychiatric harm suffered by persons connected with the incident.

The Court of Appeal considered negligence, factual foreseeability, proximity, and the limits of liability for psychiatric harm.

Legal principles:

The Court reaffirmed the application of the Spandeck framework and examined the need to establish a sufficiently close relationship between the defendant's conduct and the claimant's injury.

Application to AI infrastructure:

An AI-based conflict-prevention system might issue an incorrect emergency alert that causes a dangerous public reaction.

A claimant who suffers physical or legally recognised psychiatric injury may seek compensation if the relevant legal requirements are established.

The case is particularly useful for examining the limits of foreseeable harm and the need to establish an adequate connection between the negligent conduct and the injury.

Importance: The decision assists in evaluating the extent of recoverable injury arising from failures in automated warning and emergency-response infrastructure.

4. Who can be held liable for AI-based conflict-prevention failures?

Liability in Singapore may be distributed among several participants rather than automatically assigned to the AI developer.

A. AI developers

Developers may face claims where they fail to exercise reasonable care in:

Training and testing predictive models.

Detecting foreseeable model errors.

Designing appropriate safety mechanisms.

Preventing predictable algorithmic failures.

Communicating known system limitations.

Providing adequate technical documentation.

For example, a developer supplies a public-space conflict-detection system that consistently misclassifies particular ordinary activities as aggressive behaviour.

If the foreseeable consequences include wrongful detention, physical injury, or other legally recognised harm, the developer's conduct may be examined under negligence principles.

B. Infrastructure operators

Operators are particularly important because they control the real-world deployment of AI systems.

Potential failures include:

Ignoring repeated system warnings.

Allowing outdated predictive models to remain operational.

Failing to establish human-supervision procedures.

Using inaccurate or incomplete information.

Failing to investigate false-positive alerts.

Continuing operations after identifying serious safety defects.

An operator cannot necessarily transfer all responsibility to an AI supplier merely by arguing that the technology generated the harmful recommendation.

C. Government agencies

Where government authorities deploy AI for conflict prevention, public safety, or infrastructure management, questions of governmental responsibility may arise.

Relevant issues include:

Whether a duty of care exists.

Whether the authority exercised operational control.

Whether the harm was foreseeable.

Whether the claim concerns an operational failure or a broader policy decision.

Whether statutory protections or other public-law considerations apply.

The mere fact that a government agency uses an AI system does not automatically establish liability.

D. Joint and multiple responsibility

A single incident may involve several negligent acts.

For example:

A developer creates an unreliable predictive model.

A data provider supplies inaccurate information.

An integrator incorrectly configures the system.

An operator ignores warning notifications.

A security company acts excessively on an automated recommendation.

The court would examine the legal responsibility of each party, causation, contractual arrangements, and applicable principles governing concurrent liability.

5. Major forms of AI-based conflict-prevention infrastructure liability

5.1 Predictive policing and public safety failures

AI systems may identify locations or individuals associated with a predicted risk of confrontation.

Potential civil claims may arise from:

Incorrect identification.

Unjustified security interventions.

Physical injuries.

Negligent surveillance.

Inaccurate risk classifications.

Failure to correct known errors.

A major issue is whether the AI prediction was merely advisory or whether it directly influenced an intervention.

The stronger the operator's reliance on the system and the greater its control over the resulting action, the more important the operator's decision-making process becomes.

5.2 Smart transport infrastructure conflicts

AI-enabled traffic infrastructure can prevent road conflicts by coordinating traffic lights, identifying dangerous driving patterns, and detecting congestion.

Potential liability includes:

Incorrect traffic-signal coordination.

Failure to identify dangerous road conditions.

Delayed emergency-vehicle prioritisation.

Incorrect pedestrian detection.

Malfunctioning automated barriers.

Failure to respond to known system vulnerabilities.

A transport operator or infrastructure manager may face negligence claims if a foreseeable system defect results in an accident and the legal elements of liability are established.

5.3 Workplace conflict-prevention systems

Employers may use AI to identify workplace disputes, employee dissatisfaction, harassment risks, or potential confrontations.

Possible claims include:

Wrongful disciplinary intervention.

Unlawful disclosure of personal information.

Negligent workplace safety management.

Discriminatory treatment.

Breach of employment obligations.

For example, an AI system incorrectly classifies an employee as a workplace violence risk and the employer uses that classification to justify disciplinary action.

The resulting dispute may involve employment law, contract law, data protection, and negligence, depending on the circumstances.

5.4 Public infrastructure and environmental conflict prevention

AI systems may monitor:

Water supply.

Electricity distribution.

Drainage systems.

Public transport.

Waste-management networks.

Environmental hazards.

A failure to detect or communicate a foreseeable infrastructure risk may cause property damage, physical injury, or interruption of essential services.

The Singapore Shooting Association decision is particularly relevant to the importance of infrastructure inspection, maintenance, control, and causation.

5.5 AI-based community conflict management

Community-management platforms may predict disputes between residents, property owners, tenants, and neighbourhood groups.

Potential liability may arise from:

Incorrect allegations of threatening behaviour.

Automated restrictions on access to shared facilities.

Unjustified disclosure of personal information.

Inaccurate dispute classifications.

Failure to provide effective human review.

A private platform operator may face contractual or negligence claims depending on its relationship with affected persons and the nature of the harm.

6. Important statutory and regulatory considerations

Several Singapore legal frameworks may apply to AI-based conflict-prevention infrastructure.

A. Personal Data Protection Act

The PDPA is especially important where AI systems monitor individuals to prevent conflicts.

For instance, a commercial building uses facial recognition and behavioural analysis to predict potential confrontations.

If the organisation improperly collects, uses, discloses, or protects personal data, it may face enforcement consequences.

The organisation should implement appropriate data governance, access controls, retention policies, and security measures.

B. Cybersecurity Act 2018

An AI conflict-prevention platform connected to critical infrastructure may be vulnerable to cyberattacks.

Examples include:

Manipulation of conflict-risk predictions.

Interference with emergency communication.

Unauthorised access to surveillance networks.

Disruption of transport-management infrastructure.

Modification of automated control instructions.

Where the infrastructure falls within the Act's regulatory scope, applicable cybersecurity obligations must be considered.

C. Workplace Safety and Health Act 2006

If AI systems are used to identify workplace confrontations or dangerous conditions, employers and other relevant duty holders must consider their applicable workplace safety responsibilities.

An employer should not assume that installing an AI monitoring system automatically fulfils its safety obligations.

D. Protection from Harassment Act 2014

If an AI system generates or disseminates false or harmful information about an individual, the circumstances may raise issues under the Protection from Harassment Act.

However, the statutory requirements must be independently established. An inaccurate AI prediction is not automatically harassment.

7. Defences and limitations of liability

Defendants may rely on several arguments depending on the circumstances.

1. Absence of duty of care

A developer may argue that there was insufficient legal proximity between it and the claimant.

2. No breach of reasonable care

The defendant may establish that the system was reasonably designed, tested, maintained, and supervised.

3. Lack of causation

The defendant may argue that the harm would have occurred even without the alleged negligence.

4. Contributory negligence

Under the applicable principles of Singapore law, a claimant's own negligence may reduce recoverable damages.

5. Independent third-party intervention

A defendant may argue that an independent intervening act caused the loss. Whether this succeeds depends on the facts and applicable causation principles.

6. Contractual allocation of responsibility

A defendant may rely on contractual provisions allocating responsibilities, subject to their enforceability and the applicable legal rules.

7. Statutory protection

A relevant statutory immunity or limitation may apply in particular circumstances, but it cannot simply be assumed.

8. Remedies available to affected persons

Depending on the cause of action and the evidence, affected claimants may seek:

Compensatory damages: For legally recoverable physical injury, property damage, or financial loss.

Personal injury damages: Where negligence causes recognised physical or psychiatric injury.

Contractual damages: For loss resulting from breach of an enforceable contractual obligation.

Injunctive relief: To prevent continuing or threatened unlawful conduct where the legal requirements are met.

Declarations: To determine the parties' legal rights and obligations.

Data protection remedies: Regulatory complaints and other remedies available under applicable data protection law.

The availability of compensation depends on the particular cause of action. Not every inaccurate prediction, automated warning, or unsuccessful conflict-prevention attempt creates an entitlement to damages.

9. Practical legal framework for determining liability

A court considering a dispute involving AI-based conflict-prevention infrastructure may examine the following sequence:

Liability assessment

Identify the AI system

Determine its purpose, technical functions, users, and operating environment.

Identify the responsible parties

Establish who designed, supplied, installed, operated, and maintained the system.

Establish the duty of care

Apply the Spandeck foreseeability, proximity, and policy framework.

Examine negligence

Investigate defective design, inadequate testing, poor supervision, or failure to respond to warnings.

Establish causation

Determine whether the alleged failure caused the claimant's injury or loss.

Determine remedies and defences

Consider damages, contractual remedies, contributory negligence, and applicable statutory protections.

Evidence required in AI infrastructure litigation

Important evidence may include:

AI model documentation and technical specifications.

Training and validation records.

Algorithmic risk-assessment reports.

System logs and automated alerts.

Cybersecurity incident records.

Maintenance and inspection reports.

Human-supervision records.

Contracts between technology providers and operators.

Expert evidence concerning system failures.

Records showing how the AI recommendation influenced the final intervention.

Preserving these records is particularly important because AI systems may update their models, overwrite operational logs, or change their predictions over time.

10. Future development of Singapore civil liability law

Singapore may increasingly need to address the following issues as AI becomes integrated into public and private infrastructure.

A. Algorithmic duty of care

Courts may need to clarify the circumstances in which developers and operators owe duties to individuals affected by automated risk predictions.

B. Mandatory human oversight

High-impact AI interventions may require meaningful human supervision, particularly where decisions affect personal safety, access to essential services, or individual rights.

C. Shared responsibility

Future disputes may require clearer allocation of liability between AI developers, data providers, system integrators, infrastructure operators, and public authorities.

D. Algorithmic transparency

Access to system logs, explanations, and relevant technical evidence may become increasingly important in establishing negligence and causation.

E. Preventive infrastructure governance

Organisations may increasingly adopt independent AI audits, safety testing, incident reporting, and continuous monitoring.

F. Protection against algorithmic discrimination

AI systems should be assessed for inaccurate risk classifications and unfair treatment arising from biased data or defective models.

11. Conclusion

Civil Law and AI-Based Conflict Prevention Infrastructure Liability in Singapore represents an emerging area of legal responsibility involving negligence, infrastructure safety, contractual obligations, personal data protection, cybersecurity, and public accountability.

The most important judicial foundation is Spandeck Engineering v Defence Science & Technology Agency [2007] SGCA 37, which establishes the framework for determining negligence-based duties of care.

Other important decisions, particularly RSP Architects, See Toh Siew Kee, Tan Juay Pah, and Singapore Shooting Association, provide useful principles concerning professional responsibility, occupiers' liability, construction safety, maintenance failures, and causation.

The central principle is that deploying AI to prevent conflict does not eliminate the legal responsibility of the people and organisations controlling its design, implementation, supervision, and operation. However, liability must be established through an applicable legal cause of action rather than presumed merely because an AI system has failed.

Case-law reference note

The principal judgments discussed above are:

Spandeck Engineering (S) Pte Ltd v Defence Science & Technology Agency [2007] SGCA 37.

RSP Architects Planners & Engineers v Management Corporation Strata Title Plan No 1075 [1999] SGCA 30.

See Toh Siew Kee v Ho Ah Lam Ferrocement (Pte) Ltd [2013] SGCA 29.

Tan Juay Pah v Kimly Construction Pte Ltd [2012] SGCA 17.

Animal Concerns Research & Education Society v Tan Boon Kwee [2011] SGCA 2.

Singapore Shooting Association v Singapore Rifle Association [2017] SGHC 266.

Ngiam Kong Seng v Lim Chiew Hock [2008] SGCA 23.

These authorities establish the underlying legal principles; none should be represented as a direct judicial ruling on AI-based conflict-prevention infrastructure.