Malaysia Must Treat AI Influence Operations as a Systemic Threat

Anthropic’s 10 September threat intelligence report on a commercial influence operation targeting Malaysia should unsettle the nation’s institutions. A now-deleted Claude account helped build a system combining constituency profiling, more than 1,000 fake X accounts, the fabricated Malaysian Pulse news outlet, and dossiers aimed at political and civil-society targets. Real census and electoral data, alongside millions of voter records, targeted race, religion and royalty across all 222 parliamentary constituencies.

Malaysia is familiar with cyber-trooper politics. What differs here is the infrastructure, which brings together voter data, synthetic identities, AI-assisted content production, artificial engagement and operational dashboards assembled into a commercial service. That makes the case more than another disinformation campaign. It is an informational systemic risk.

That distinction matters. The disclosure does not show that a Malaysian election was compromised. Anthropic did not identify a client or tie the operation to a particular contest. It found no evidence that the actor’s pursuit of a contract with Malaysia’s communications regulator succeeded or that the network broke into authentic online communities, making it Category Two on the Breakout Scale, a metric of threat severity proposed by the Brookings Institute. Although the operation did not affect an election, the revelation that such infrastructure was assembled for sale should cause grave concern.

The platform’s design matters more than any single post or account. Anthropic uncovered clear links to BBS Bilisim Teknolojileri, an Istanbul technology company, which it said sold access to the platform as a “paid influence-as-a-service capability”. BBS has denied involvement, knowledge or approval and is examining whether its name or digital assets were used without authorisation. The attribution is contested; neither Anthropic’s findings nor BBS’s denial establishes a Malaysian sponsor.

The case raises an urgent question about how a commercial system can obtain and operationalise millions of voter records for constituency-level political targeting. Claude Code also managed fake accounts and artificial engagement to simulate popularity.

The authorities must be proactive and vigilant in tackling the root problem: corruption of the information environment.

Anthropic observed a request to generate one million artificial views on the sitting prime minister’s account, although the figures could not be independently verified and the report does not link the request to the prime minister or his office.

The operation extended to information production. Malaysia Pulse, the now-defunct counterfeit news outlet, scraped Malaysian reporting and rewrote it under invented bylines, republishing material from Russian and Chinese state-aligned outlets without attribution that would allow readers to judge its origins. The operation also used fabricated dossiers to make allegations against an opposition politician and civil society organisations. By circulating false claims through polished and credible-looking reports, the scheme distorted the signals that journalists, political parties, regulators and voters use to decide what is real, urgent, or popular.

The informational Systemic Risk (ISR) framework explains how degraded or manipulated information flows can create risks for interdependent social, political, and institutional systems (Figure 1). Voter profiling determines whom to target, synthetic accounts manufacture apparent public responses, artificial engagement simulates popularity, pseudo-news supplies a journalistic wrapper and fabricated dossiers inject allegations with false authority. These components can reinforce one another. The issue is not whether every voter believes a false post, but whether coordinated systems blur the distinction between authentic and engineered signals across politics, public agencies, news media, civil society, platforms and business.

Figure 1. How AI Influence can Create Informational Systemic Risk in Malaysia

If the risk is systemic, the response cannot rest with one regulator alone. The Malaysian Communications and Multimedia Commission (MCMC) has said it is examining Anthropic’s report while stressing that it does not engage in the implicated activities or arrangements. MCMC’s review should trigger a wider inter-agency inquiry into whether Malaysian data, intermediaries or clients were involved. MCMC should coordinate with platforms to preserve relevant accounts, domains and records; the Election Commission (EC) should establish whether voter data came from public electoral rolls or another Malaysian source; and relevant enforcement agencies should investigate any unlawful acquisition or use of personal data.

That institutional gap is mirrored in Malaysia’s legal framework. Malaysia’s Online Safety Act 2025 requires licensed providers to mitigate exposure to harmful content and sets grounds and mechanisms for making harmful content inaccessible, but it does not establish an election-specific standard for coordinated inauthentic behaviour (CIB). CIB should be recognised as a distinct electoral harm that engages in deceptive coordination and undisclosed, networked operations. This requires law enforcement to go beyond monitoring the veracity of online posts and taking down those that meet the falsity threshold. The authorities must be proactive and vigilant in tackling the root problem: corruption of the information environment.  

Policy Implications

That policy gap is no longer abstract. Before the next general election, MCMC and EC should establish a standing integrity protocol with platforms, web hosts and frontier model providers to preserve evidence and share verified indicators of false domains, coordinated account behaviour, impersonation and attempts to bypass model safeguards. Large platforms should also stress-test whether recommendation systems, advertising products, targeting tools and account controls can be manipulated to inflame fault lines including ethnic, religious and political divisions. Public agencies should be prohibited from purchasing undisclosed political engagement or influence operation services. These measures address the machinery of manipulation without turning election integrity into a censorship mandate.

The influence operation crossed jurisdictions before Anthropic intercepted it. As host of the ASEAN AI Safety Network Secretariat, Malaysia can help build a regional process for sharing verified indicators and coordinating evidence-preservation requests when cross-border operations emerge. Anthropic identified the account through internal detections and removed it. Malaysia’s task is to prevent another network exploiting the same gaps before it reaches genuine communities and undermines public institutions.

2026/277