AI Agents Face New Wave of Social Engineering Attacks

The report highlights a dangerous evolution in cybersecurity tactics: after a decade of mastering human manipulation, threat actors are now learning to socially engineer AI. This transition has led to a surge in high-profile incidents, ranging from data exfiltration and agent hijacking to denial-of-wallet attacks that inflate operational costs without needing to breach private records.

To counter these vulnerabilities, OrcaRouter is releasing a suite of security tools at no additional cost. The new Firewall and Guardrails features operate directly within the OrcaRouter AI gateway, allowing organizations to monitor prompt inputs and restrict agent actions—such as tool calls, network access, and spending limits—without requiring changes to existing application code. By screening for jailbreaks and policy violations, these tools aim to secure the agents that now act on behalf of modern enterprises.