insights › tag
ai-security
2 posts tagged ai-security on prest0 insights.
- Beyond pass@1: Engineering Closed-Loop Verification for Production AI Agents — Why AI agents pass single-run demos (pass@1) but fail in production—and how a four-stage closed-loop verification architecture delivers repeatable pass^k reliability.
- Agentic Memory Security and Privilege Isolation: Stopping Context Poisoning in Production AI — How engineering teams stop indirect prompt injection, memory poisoning (OWASP ASI06), and privilege escalation using capability-isolated control planes and scoped memory.