New Archestra's OpenAPPA Saturates Two Major Security Benchmarks with a 0% Attack Success Rate
0%. Archestra's OpenAPPA stopped every attack in enterprise agent benchmarks—while Claude Code auto mode and Microsoft FIDES leaked data 10% and 31% of the time.
Why it matters
A new open-source security engine claims zero successful prompt-injection and hallucination-driven exfiltration attacks on two production-grade agent benchmarks. The comparative data (vs. Claude Code auto mode and FIDES) is specific and testable, making this relevant to practitioners evaluating agent safety in multi-step workflows.
The key facts
6 to knowArchestra released OpenAPPA, an open-source security engine for agent data exfiltration prevention
Claimed 0% attack success rate on Bench-Corp (20 multi-step enterprise workflows) and AgentThreatBench
Claude Code auto mode: 10% attack success rate on same benchmarks
Microsoft FIDES: 31% attack success rate on same benchmarks
Attack vectors: prompt injection and model hallucination-driven exfiltration
Published by InfoQ, no independent third-party validation cited
Go to the source
InfoQ AI/MLinfoq.com
Publisher excerpt: Archestra released OpenAPPA, an open-source security engine designed to stop data exfiltration caused by prompt injection or model hallucination. The team reports zero successful attacks when running security benchmarks Bench-Corp (20 multi-step enterprise workflows) and AgentThreatBench, versus…