Meta ramps up fight against AI-driven scams in Asia-Pacific

Meta has intensified its regional crackdown on online scams following the release of its latest Adversarial Threat Report, which highlights the rising sophistication of cybercrime syndicates across the Asia-Pacific region.

These criminal networks, increasingly driven by artificial intelligence, have adopted scalable tactics such as impersonation fraud, scripted outreach and deepfake media to deceive users and defraud victims.

A major focus of Meta’s latest report is the coordinated disruption of networks based in Cambodia. These operations impersonated police and government officials to defraud users in Australia, Indonesia, the Philippines, Thailand, Vietnam, and Singapore. 

According to Meta, such scams have become increasingly cross-border and scalable, leveraging automation and encrypted messaging platforms to conceal identities.

Targeting victims twice: ‘recovery scams’ on the rise

A common scam archetype identified in the report involves criminals posing as law enforcement or regulatory agencies offering to recover funds lost to previous scams—for a fee.

These “double-tap” scams exploited Facebook groups and online forums where previous scam victims sought help. Fraudsters used fake logos, official-sounding names such as “Cyber Crimes Investigations Service”, and images of uniformed officers to lend credibility to their outreach.

Once engaged, victims were persuaded to switch to encrypted apps and asked to pay upfront for investigative costs or refunds that never materialised.

Meta’s investigations linked many of these impersonation networks to Cambodian-run scam centres, dismantling over 6,400 associated Facebook accounts and pages between January and October 2025.

AI deepfakes and impersonation: a rising threat

The report also highlights the growing use of artificial intelligence in scams across the region. Techniques include:

• Deepfake videos of public figures used to promote fake investment opportunities;

• AI-generated text for mass-personalised outreach;

• Identity spoofing of celebrities and government officials in scam advertisements.

These trends are especially visible in Australia, India and Singapore, where impersonation scams are among the most commonly reported threats.

Meta has responded by integrating new safeguards such as the Llama Firewall, the Rule of Two framework to secure AI agents, and enhanced behavioural detection tools to identify suspicious conversations. The company’s facial recognition programme now covers nearly 500,000 public figures, aimed at stopping the misuse of their likeness in fake ads.

Meta’s four-pillar approach to scam mitigation

To address the rise in adversarial threats, Meta has adopted a global strategy built around four pillars:

1. Platform resilience: Hardening defences using AI detection, facial recognition, and policy updates to address new threats.

2. User empowerment: Providing users with scam alerts, safety education, and enhanced reporting tools. New features in Messenger and WhatsApp include proactive scam warnings and AI-assisted chat reviews.

3. Disruption and deterrence: Meta removed over 134 million scam-related advertisements in 2025 and disabled nearly 12 million accounts tied to criminal scam centres. Reports of scam ads dropped by 50% over the last 15 months.

4. Cross-society collaboration: Meta works with law enforcement, financial institutions, and other platforms to share intelligence. In Singapore, cooperation with the police led to arrests linked to illegal online gambling and scams, with victims recovering over US$175,000.

A regulatory footnote: Singapore’s first enforcement directive under OCHA

While Meta’s actions are part of a long-standing global security programme, recent developments in Singapore underscore the rising regulatory scrutiny surrounding impersonation scams.

On 24 September 2025, the Singapore Police Force issued Meta an Implementation Directive under the Online Criminal Harms Act (OCHA). This legally binding order compels Meta to:

• Enhance its facial recognition measures for Singapore-based users; and

• Prioritise scam-related user reports that involve impersonation of Government Office Holders on Facebook.

This marked the first directive issued under the Act, and follows the Ministry of Home Affairs' observation of over 2,000 impersonation scam assets on Facebook between June 2024 and June 2025. Meta must comply by 30 September 2025 or face fines of up to S$1 million, plus S$100,000 for each day of continued non-compliance.

While not the cause of Meta’s broader actions, this directive reflects growing expectations placed on platforms to prevent online harm and protect institutional trust.

Outlook: shared responsibility in a fragmented threat landscape

Meta continues to expand its technical capabilities and cross-sector cooperation to meet the scale of adversarial online threats. As AI enables scammers to scale their operations with unprecedented efficiency, companies, regulators, and users alike must adapt in step.

The company says it is committed to publishing threat signals, working with law enforcement, and improving detection systems as part of a “whole-of-society” approach. In regions like Southeast Asia, where regulatory frameworks such as OCHA are emerging, this collaboration may increasingly define the digital security landscape.