ASEAN Faces Risks from Cross-Border AI Agent Swarms

Lowy Institute reports that the emergence of autonomous, agentic AI agents creates new forms of online environments where software systems interact and generate content at scale. The article identifies an episode involving the acquisition of Moltbook, described as an autonomous social network of AI agents, and reports that approximately 1.5 million authentication keys and identities were exposed in that incident, potentially enabling attackers to manipulate agents to extract or delete sensitive data, according to Lowy Institute. The piece states that agentic AI can send messages in a user's name and initiate actions with consequences, and that swarms of agents can amplify disinformation and automate cybercrime across national borders, per Lowy Institute.

Industry-pattern observations: Agentic systems differ from classical human-directed AI primarily in autonomy and delegated authority. That change reduces the human-in-the-loop friction that previously slowed malicious campaigns, enabling faster content generation, coordinated amplification, and automated probing of vulnerabilities. For practitioners, this elevates the importance of identity hygiene, key management, and monitoring of agent-driven activity across interfaces.

Industry-pattern observations: Southeast Asia's linguistic diversity, fragmented regulatory regimes, and varied digital infrastructure make coordinated detection and mitigation harder than in more homogeneous markets. Cross-border propagation is likely to outpace national rule-making unless regional cooperation mechanisms are adapted. Lowy Institute frames the problem as a governance mismatch between existing laws designed for human actors and emerging agentic behaviour.

• •Signals of large-scale credential exposure or API key leaks linked to agent platforms.
• •Policy moves by ASEAN member states or regional bodies that extend liability, transparency, or auditing requirements to autonomous software.
• •Technological controls adopted by platforms, such as agent sandboxing, fine-grained permissioning, and provenance tagging of agent-originated content.

For practitioners: Security, platform, and policy teams in and beyond Southeast Asia should prioritise detection pipelines for automated, coordinated agent activity and strengthen secrets management and agent permission models. Observers should also track whether regional regulators propose cross-border cooperation frameworks that explicitly reference autonomous software.