90% of Successful Attacks Seen in the Wild Resulted in Leaked Sensitive Data

Pillar Security’s press release highlights findings from its “State of Attacks on GenAI” report analyzing telemetry from 2,000+ production AI apps: 20% of jailbreak attempts bypass guardrails, 90% of successful attacks leak sensitive data, attacks complete in ~42 seconds with ~5 interactions, and prompt injections are increasing in frequency and sophistication; it notes common techniques (ignoring prior instructions, Base64-encoded prompts), attacker motives (data theft, disinformation, phishing, malicious code), and urges adaptive, real-time AI security over static controls as organizations adopt copilots and autonomous agents.