Artificial intelligence firm Anthropic disrupted a covert influence operation that targeted voters in Malaysia, marking a significant intervention by a major tech company against automated political manipulation in Southeast Asia. According to an incident report published by Anthropic, the threat actors used the company’s Claude large language models to generate deceptive content, translate political narratives into local languages, and scale social media accounts designed to sway public opinion.
How the Operation Used Claude Models
The threat actors leveraged Anthropic’s AI systems to automate the creation and distribution of politically sensitive text. According to Anthropic’s threat intelligence team, the campaign utilized Claude to draft social media posts, refine messaging for specific demographic groups, and translate materials into multiple languages spoken in Malaysia. By automating these tasks, the operators managed to produce high volumes of localized propaganda without maintaining a large team of human writers. Anthropic stated that the accounts involved in the activity violated the company’s acceptable use policies, which explicitly prohibit the deployment of AI models for deceptive campaigns, voter suppression, or unauthorized political lobbying.
Detection and Disruption Timeline
Anthropic’s security team detected the anomalous activity patterns during routine monitoring of model usage and API traffic. Upon identifying the cluster of accounts tied to the influence campaign, the company revoked API access and terminated the user profiles associated with the operation. While threat actors frequently attempt to bypass safety filters by obfuscating their prompts, automated classifiers and human reviewers flagged the coordinated nature of the political messaging. The intervention prevented the operators from continuing to use Claude for generating further batches of election-related content.
Broader Implications for Regional Security
The disruption highlights growing concerns among cybersecurity experts regarding the accessibility of generative AI for state and non-state actors seeking to interfere in democratic processes. Malaysia has experienced increased scrutiny over online disinformation, particularly on platforms like TikTok, Facebook, and X. Security analysts note that commercial AI tools lower the technical barrier to entry for influence operations, allowing small groups to project outsized influence across multilingual digital spaces. Regulatory bodies and technology companies continue to grapple with balancing open access to advanced models against the risk of malicious exploitation during sensitive political events.
Worth a look