\n
Singapore\u2019s scam landscape looked better on paper in 2025 than it had in years. Total scam and cybercrime cases<\/u> fell by 24.8% to 41,974 in 2025, and funds lost dropped from S$1.1 billion in 2024 to S$913.1 million<\/b>. But beneath that headline figure, a more concerning trend was visible, where the median loss per case rose from S$1,389 in 2024<\/b> to S$1,644 in 2025<\/b>.\u00a0<\/p>\n<\/p>\nTable of contents<\/p>\n
<\/default:svg><\/button><\/p>\n\n1. Phishing remains a key entry point, but AI enhances its effectiveness.<\/li>\n 2. AI-powered scam campaigns are scaling and becoming harder to contain.<\/li>\n 3. PromptLock: A proof of concept with real consequences<\/li>\n 4. ESET\u2019s upcoming AI defence\u00a0<\/li>\n 5. The human element: AI tools are not a replacement<\/li>\n 6. Practical Steps for Singapore Businesses<\/li>\n<\/ol>\n<\/section>\nFor Pamela Ong, Country Manager of Singapore and Asia at ESET, that divergence between volume and severity was not a contradiction. It was a signal.<\/p>\n
\n
Play<\/span><\/button><\/div>\n<\/div>\n\u201cWhat we\u2019re seeing is not necessarily a reduction in attacker activity, but a change in approach,\u201d Ong said in an email interview with HardwareZone on the sidelines of Gitex AI Asia 2026. \u201cInstead of relying purely on volume, attackers are focusing more on precision. They are using better reconnaissance, more convincing impersonation, and increasingly AI-generated content to target victims more likely to yield higher returns.\u201d<\/p>\n
Her comments came as ESET showcased its upcoming AI protection capabilities at Gitex AI Asia 2026, held in April in Singapore. The new features, first demonstrated at RSAC 2026 and set to launch later this year, were designed to address a threat landscape that Ong described as evolving faster than most organisations were prepared for.\u00a0<\/p>\n
Phishing remains a key entry point, but AI enhances its effectiveness.<\/h2>\n\nPamela Ong, Country Manager of Singapore and Asia at ESET<\/p>\n
Photo: HWZ<\/p>\n<\/figcaption><\/figure>\n
Ong was careful to point out that the fundamentals of the threat landscape had not changed as dramatically as some of the technology headlines might suggest. \u201cMost successful breaches still begin with socially engineered emails and malicious links,\u201d she said. ESET\u2019s own data showed that phishing remained the leading cause of corporate breaches in Singapore, accounting for close to a third of detected threats. \u201cThis tells us attackers are still relying on familiar entry points but are executing them more effectively.\u201d<\/p>\n
What had changed substantially was the quality of those attacks, and AI was the reason. In the first five months of 2025, a third of phishing emails were found to contain high volumes of text<\/b>, indicative of large language model generation. Ong said the implication for how people assess whether a message was legitimate was significant and largely uncomfortable.<\/p>\n
\u201cLarge language models have made it much easier for attackers to produce well-written, coherent, and contextually relevant messages at scale. As a result, many phishing emails today look professional and credible, which removes one of the most obvious warning signs that people previously depended on.\u201d<\/p>\n
The practical advice she offered was a shift from visual assessment to behavioural interrogation. \u201cInstead of focusing on how a message looks, it\u2019s more important to focus on what it is asking you to do. Is it creating urgency? Is it asking you to click a link, download an attachment, or share sensitive information? Does the request match the sender\u2019s usual behaviour?\u201d Verification, she said, was now the more reliable defence, not grammar or formatting.<\/p>\n
AI-powered scam campaigns are scaling and becoming harder to contain.<\/h2>\n ESET\u2019s H2 2025 Threat Report<\/u> had flagged a range of AI-assisted threats, including the Nomani scam network, which used AI-generated videos to spread fraudulent content across multiple platforms simultaneously. Ong described the scale of automation behind these campaigns as one of the more striking findings in recent threat intelligence.<\/p>\n
\u201cWhat is notable about campaigns like Nomani is not just the use of AI-generated content, but how it allows attackers to scale distribution across multiple platforms quickly, making scams harder to contain and more persistent,\u201d she said. ESET observed over 159,000 detections tied to a single phishing domain, usrpubtrk[.]com, which Ong said signalled that attackers were relying on highly automated infrastructure to deploy and rotate campaigns at speed.<\/p>\n
On whether defenders were keeping pace, her assessment was measured rather than reassuring. \u201cAttackers are able to iterate very quickly. The organisations that are keeping up are those who added some form of AI on the defensive side as well, for example, through behavioural detection and threat intelligence. Those relying only on traditional indicators or manual processes will find it increasingly difficult to keep up.<\/p>\n
PromptLock: A proof of concept with real consequences<\/h2>\n\nSome of the crowd at ESET\u2019s booth at Gitex Asia 2026<\/p>\n
Phto: ESET<\/p>\n<\/figcaption><\/figure>\n
One of the more significant findings from ESET\u2019s H2 2025 Threat Report was the identification of PromptLock, described as the first known ransomware strain to leverage AI in its attack process, capable of encrypting, exfiltrating or destroying data. Ong carefully placed the discovery in context.<\/p>\n
\u201cPromptLock highlights where this is heading: towards greater automation across different stages of an attack, from identifying targets, to evading detection, and potentially adapting in real time,\u201d she said. \u201cThis could make attacks faster and give defenders less time to respond.\u201d<\/p>\n
She was equally careful not to overstate the immediate threat. <\/p>\n
\n
\u201cMost successful attacks today still rely on more established methods like phishing and credential theft. These remain the primary ways attackers get into organisations.\u201d<\/p>\n<\/div>\n
But on the trajectory the discovery represented, she was unambiguous. \u201cWhile ESET considers PromptLock a proof of concept, the threat it represents is very real. It is a clear sign that attacks are becoming more automated and more adaptive, and organisations need to be prepared for that shift.<\/p>\n
ESET\u2019s upcoming AI defence\u00a0<\/h2>\n\nA demo of ESET\u2019s upcoming security platform for AI agents<\/p>\n
Photo: ESET<\/p>\n<\/figcaption><\/figure>\n
Against that backdrop, the features ESET showcased at GITEX AI ASIA 2026 were designed to address risks that had emerged specifically from the rapid integration of AI tools into everyday business workflows. The new capabilities, set to expand the ESET PROTECT Platform, was centred on three areas.<\/p>\n
The first was a secure browser technology that intercepted AI interactions and analysed both prompts and responses in real time. As AI tools became embedded in everyday workflows, many employees were using open cloud chatbots without IT oversight, creating what ESET described as \u201cshadow AI\u201d risks, and inadvertently exposing sensitive data, including internal documents, API keys and credentials. The new feature would flag malicious URLs submitted through:\u00a0<\/p>\n
\nchatbot prompts,\u00a0<\/li>\n detect prompt injection attempts\u00a0<\/li>\n block or monitor the upload of sensitive data in accordance with an organisation\u2019s policies.<\/li>\n<\/ul>\nThe second area was endpoint security capabilities designed to detect AI supply chain attacks. Ong explained in plain language what these attacks involved. \u201cAn AI supply chain attack happens when attackers compromise the building blocks behind an AI system, such as third-party models, plug-ins, or agent frameworks, rather than targeting the organisation directly. If one of these components is tampered with, that risk is effectively passed on to every organisation using it.\u201d<\/p>\n
ESET had already identified cases involving trojanised components in widely used AI libraries, specifically naming LiteLLM as an example, as well as autonomous agents like OpenClaw that could execute actions on a system with limited oversight. \u201cThis is becoming more relevant as organisations adopt AI tools quickly and often integrate external components into their workflows,\u201d Ong said. <\/p>\n
\n
\u201cIn many cases, these tools are introduced faster than they can be fully assessed, which creates gaps in visibility and control.\u201d<\/p>\n<\/div>\n
For Singapore businesses in particular, where AI adoption was accelerating across productivity, customer engagement and automation, she framed the message directly: \u201cAI needs to be treated as part of their broader supply chain risk.\u201d<\/p>\n
The third element was the ESET AI Skills Checker, already available as a free public tool open to non-ESET customers. Built on the same technology as ESET\u2019s endpoint security products and ESET LiveGuard, the checker analysed AI skills and agents for hidden instructions, malicious code and risky behaviour using multilayered inspection and cloud-based sandboxing. It was currently also available as a built-in feature for existing ESET Endpoint users.<\/p>\n
Ong was asked whether there was a risk that organisations over-relied on AI security tools at the expense of human expertise and security culture. She said it was a concern she was already seeing manifest in the market.<\/p>\n
\u201cWhile AI is extremely valuable in helping security teams manage scale, reduce noise, and detect threats earlier, it is a complement to human expertise, not a replacement for it,\u201d she said, adding: <\/p>\n
\n
\u201c\u201cMany successful attacks still come down to very fundamental issues, such as phishing, compromised credentials, or gaps in processes, rather than failures in advanced tools.\u201d\u201d<\/p>\n<\/div>\n
The approach she recommended was explicit about the balance required. \u201cThis involves investing in people and processes, including training, awareness, and incident response, while at the same time incorporating AI tools to strengthen detection and reduce response times. Organisations that rely too heavily on either side risk creating gaps in their overall security posture.\u201d<\/p>\n