Liability: who is responsible when AI produces false statements?
Artificial Intelligence (AI) is rapidly transforming the financial services sector, offering unprecedented opportunities - but also new legal challenges.
For FAnews readers, understanding the intersection of AI, defamation, and liability is critical as generative AI tools become embedded in operations and client interactions.
FAnews spoke to legal experts Anja Hofmeyr, Director in the Dispute Resolution practice at Cliffe Dekker Hofmeyr (CDH), and Safee-Naaz Siddiqi, Senior Associate in the firm’s Knowledge Management team, about the legal implications of AI-generated content, particularly relating to defamation risk and reputational exposure in South Africa.
Lessons from a US case
A landmark US case, Walters v OpenAI, underscores the real legal risks posed by AI “hallucinations” - AI-generated statements that are false but presented as fact. As Hofmeyr explains, “A journalist used ChatGPT for research. ChatGPT hallucinated an assertion that Mark Walters, an American public figure, had embezzled funds. The journalist verified that the assertion wasn’t accurate, but Walters still sued OpenAI, the developer of ChatGPT, for the defamatory content hallucinated by ChatGPT.”
While the defamation claim against OpenAI failed due to specific circumstances, the case reveals how easily AI misinformation can spread. “Had the journalist published the assertion without verification, he would likely have faced his own defamation lawsuit - one wherein he would not be able to rely on OpenAI's defences. He avoided a suit only because he exercised professional judgment and fact-checked the AI's output,” Hofmeyr says.
Siddiqi adds, “The value of the Walters case isn’t in the outcome, but in what it reveals about the AI landscape. AI platforms routinely generate false, damaging statements about real people. The only protection is scrutinising the AI-generated output, exercising sound judgment and verifying its accuracy before relying on it.”
For South African businesses, this case serves as a cautionary tale. Relying blindly on AI outputs without verification could lead to costly defamation claims and reputational damage.
Are defamation laws AI-ready?
The legal framework for defamation in South Africa traditionally assumes human agency - an author or publisher responsible for the content. Hofmeyr notes, “The elements of defamation in terms of South African law - the wrongful and intentional publication of a defamatory statement concerning another, which causes financial or reputational harm - is usually premised on human agency.”
However, AI-generated content challenges this assumption. “Large Language Models, by virtue of their nature, do not involve human intervention in generating content, so some of the elements or defences are not directly applicable when it comes to machine-generated content,” Hofmeyr explains.
Siddiqi elaborates, “The publication requirement traditionally involves making the defamatory content known to others; with AI, the fact that the output is generated autonomously and disseminated instantly would probably suffice to satisfy this requirement.”
Courts would likely presume defamatory meaning and unlawful intent once defamatory content is identified. However, “these presumptions should be rebuttable on the basis that a Large Language Model (or its developer) lacks the required intention to defame - at least at present,” Hofmeyr suggests.
Drawing from the Walters case, she proposes that courts consider whether AI developers took “reasonable steps to prevent their LLM from hallucinating, and whether disclaimers and robust warnings were not only included in terms of use but specifically drawn to the attention of users.”
“If the developer can show reasonable steps taken to minimise hallucinations, it may rely on a defence akin to the reasonable-publication defence available for traditional media,” she says. Similarly, users relying on AI-generated content must show they acted reasonably - such as by fact-checking - to avoid liability.
As it stands, without judicial or statutory development, existing laws may not fully address AI’s unique challenges, signalling the need for evolving legal frameworks.
Liability: who is responsible?
When AI produces false or damaging statements, determining liability is complex. “Under South African law, liability generally attaches to the party that publishes or repeats the defamatory material,” Hofmeyr says.
She warns, “If someone republishes an AI-fabricated statement without reasonable verification, they will likely be held liable for defamation, as the presumption of unlawfulness and intention will apply, and there would not be a basis to rebut the presumptions.”
What about the AI developers themselves? Hofmeyr advises courts should consider whether the developer acted reasonably under the circumstances. “This may involve continuous development to reduce risks, user warnings, and implementation of industry-standard safeguards,” she says. Liability will ultimately depend on the specific facts of each case.
Siddiqi notes the involvement of other actors, such as hosting platforms, prompt engineers, or API integrators, “may also be relevant to liability questions, though their responsibility is less clear at this stage.”
For businesses, this underscores the critical importance of governing AI use internally to reduce risk and prevent reputational harm.
Verifying AI-generated content
AI fabricated statements are not just technical glitches - they carry serious professional and legal consequences. Siddiqi highlights, “Legal professionals have faced sanctions for citing non-existent cases generated by generative AI tools like ChatGPT, but any professional using these tools is equally vulnerable.”
The challenge lies in AI’s sophistication. “Unlike obvious errors, AI-generated misinformation often appears credible, complete with plausible details and authoritative tone,” she says.
This makes verification protocols essential. Hofmeyr gives practical examples: “Consider a doctor relying on AI-generated treatment recommendations without verification, potentially endangering patient safety and facing medical malpractice claims. Similarly, a journalist publishing AI-generated quotes or statistics without fact-checking risks defamation lawsuits and professional credibility damage.”
In both cases, AI-fabricated statements can cascade into real-world harm when verification is absent. Safeguards must include thorough fact-checking, transparency about AI’s role, and robust human oversight.
Siddiqi stresses, “Businesses should invest in clear AI governance frameworks that outline accountability, permitted uses, disclosure mechanisms, verification protocols, and mandatory human oversight.”
Mitigating reputational risk
To mitigate reputational risks, companies must adopt layered strategies and invest in governance. “AI literacy training for employees, deploying enterprise-level AI tools with robust security and compliance, and clear accountability chains are essential,” Siddiqi says.
She also recommends avoiding reliance on freely available consumer AI tools for sensitive or regulated work, as these often lack necessary safeguards.
Senior business leaders need to prioritise their own AI literacy and actively manage exposure. Hofmeyr warns, “Proper AI governance can come with substantial costs upfront, but it’s critical for sustainable risk management.”
SA’s legal frameworks must evolve rapidly
Both experts agree that South Africa’s legal frameworks must evolve rapidly. Hofmeyr points to recent controversies, such as Meta AI’s chatbot policies “engaging children in romantic or sensual conversations,” to illustrate “the urgent gaps in existing regulation.”
Siddiqi observes, “Current laws designed for human actors struggle to address AI systems that operate at unprecedented speed and scale, interacting with millions daily.”
This regulatory void risks unchecked harmful AI behaviour and legal uncertainty for businesses and citizens. Internationally, the EU AI Act and US policy debates indicate a shift toward risk-based AI governance.
Hofmeyr concludes, “A calibrated legislative response, akin to Europe’s Artificial Intelligence Act, could provide necessary protections and clarity. A clear legal framework encourages responsible innovation while protecting citizens’ legal rights - its absence creates uncertainty, and without guardrails, often has the opposite effect.”
Writer’s thoughts
As AI tools become integral to business operations, understanding the evolving landscape of defamation risk and liability is paramount. Companies must prioritise robust AI governance frameworks, implement verification protocols, and stay attuned to rapid changes in legal regulations to safeguard against reputational damage and legal pitfalls. Please comment below, interact with us on X at @fanews_online or email me your thoughts.
