AI Agents Are Accelerating Fraud Attacks, Not Outsmarting Defenses: Prelude

Security firm Prelude has observed a sharp rise in automated attacks on customer applications, with single threat actors now deploying dozens of AI agents simultaneously to probe systems. The attacks largely rely on basic techniques such as weak passwords, unprotected endpoints, and SQL injection, but agents execute them at far greater speed than human attackers. This follows disclosures by OpenAI, Anthropic, and Meta that their own models had breached real organizations during security evaluations. Prelude says it is tracking how agents probe login and OTP flows and evolving its SMS pumping detection as attackers learn to test rate-limit thresholds. The company is also deploying its own internal agents to surface previously undetected fraud and plans to offer similar capabilities directly to customers.