As we move into the final phase of this series, these two supplementary articles address the critical “Safety and Oversight” layers of AI adoption. One focuses on the risks of the output (Defamation), while the other focuses on the responsibility of the leadership (Board Governance).
As Singaporean SMEs leverage Generative AI to scale their content—from automated social media responses to AI-drafted industry reports—they enter a high-stakes legal arena where the speed of production can outpace the speed of fact-checking. In the eyes of Singapore’s Defamation Act, a “hallucinated” falsehood generated by a machine carries the same legal weight as a lie written by a human.
The “Publisher” Trap
Under Singapore law, liability for defamation rests with the “publisher.” If your SME uses an AI tool to generate a LinkedIn post that incorrectly attributes a financial scandal or a professional failure to a competitor, your company is the primary publisher. The defence of “innocent dissemination” is notoriously difficult to claim in a commercial context where the business has direct control over the AI prompt and the final posting.
Crucially, the Criminal Law (Miscellaneous Amendments) Act 2025 and the Online Safety (Relief and Accountability) Act have sharpened the state’s focus on “inauthentic material.” While these often target deepfakes, the principle remains: if your AI-generated content causes harassment, alarm, or distress to a third party, your business may face not only a civil lawsuit for defamation but also potential regulatory action under the Online Criminal Harms Act.
The Risks of “Synthetic” Representation
For service-sector SMEs using AI-generated digital replicas or “brand voices,” the risks are even more nuanced. Using a synthetic voice that sounds too similar to a real individual without their consent can lead to claims of “tortious passing off” or defamation if that voice is used to say something the individual would never endorse. Also, while IP law is still catching up, the courts are increasingly protective of an individual’s “digital likeness.”
Practical Mitigation: The “Fact-Check First” Protocol
- Mandatory Human Editing: Never allow an AI to post directly to a public platform. Every piece of AI-generated content must pass through a “Human Truth Filter” to verify names, dates, and accusations.
- Clear Attribution: While not a legal “shield,” disclosing that content was AI-assisted can manage audience expectations. However, it does not absolve you of the duty to be accurate.
- Adversarial Prompting: Before deploying an automated response bot, test it with “trap” questions to see if it can be coerced into making defamatory statements about your competitors.
This article is part of the “AI, the Law and the SME” series by OTP Law Corporation. At OTP Law Corporation, we specialise in helping SMEs navigate the intersection of technology and law. If you are reviewing any of your commercial contracts, drafting an AI governance policy or have any other legal needs, contact us to ensure your business is protected.
