GuardAI sits between your application and your LLM provider, evaluating both prompts and responses against your organization's specific rules. When content violates a policy, GuardAI rewrites it to be safe and aligned before it reaches the user — no model retraining required. On WildGuardTest, GuardAI achieves 60% lower unsafe accept rate than the next-best system (15.8% vs 39.9%). See full benchmark results.
We test GuardAI against the same public benchmarks used in academic research. Full methodology and data.
GuardAI evaluates both prompts and responses against your organization's specific policies — not just generic safety rules. Here is how it works with a concrete example.
GuardAI detects the policy violation and rewrites the response to be safe and aligned before it reaches the user. Built-in model safety would not catch this because it is an organization-specific rule, not a generic safety concern.
Google's built-in safety lets 39.9% of unsafe content through on WildGuardTest; prompt-only classifiers like OpenAI Moderation miss 18-54%. GuardAI's multi-layer policy firewall cuts that to 15.8% on WildGuardTest and refuses 99.7% of adversarial jailbreaks — because each layer catches what the others miss.
Evaluates incoming prompts against your policy rules using semantic intent detection, catching role-playing attacks and obfuscated requests that keyword filters miss.
Policy-aware prompt evaluationAssigns a risk level to each request so that routine traffic passes with minimal latency while suspicious inputs get deeper inspection.
Adaptive inspection depthChecks every model response against your policies before delivery, flagging cipher patterns, abnormal structures, and content that violates your rules.
Response-level policy enforcementThe system adapts to new attack patterns as they emerge, updating detection models without requiring changes to your integration.
Live learningThe Jailbreak Against Moderation (JAM) attack demonstrates why multi-layer defense is essential. JAM attacks achieved a 75.17% average success rate (19.88x higher than baselines) and only 10.21% filtered-out rate across four leading commercial and open-source models.
Collect harmful texts and pass through moderation API to obtain harmfulness scores.
Fine-tune model to mimic guardrail behavior, enabling adversarial optimization.
Use greedy coordinate descent to find character sequences that reduce detection scores.
Combine optimized ciphers with malicious prompts to bypass both input and output guardrails.
A 4-role collaborative LLM system that tests guardrail effectiveness against natural-language attacks.
Converts government safety guidelines (EU AI Act, NIST) into specific test questions that should violate policies.
Guideline to test casesCreates natural-language "playing scenarios" by reorganizing jailbreak patterns from a knowledge graph of 78+ techniques.
Knowledge graph to scenariosMeasures jailbreak success using semantic similarity scores between target LLM responses and expected safe refusals.
Response to success scoreAnalyzes failed attempts and suggests iterative improvements to reduce similarity scores (max 10 iterations).
Feedback loop to generatorPublished research shows 75-87% attack success rates against unprotected models. These are the threats GuardAI was purpose-built to stop.
JAM attacks achieved a 75.17% average success rate — 19.88x higher than baselines — with only a 10.21% filtered-out rate, roughly 1/6 lower than the baseline rate of 54.76%. Testing covered four leading commercial and open-source models, confirming the threat is systemic rather than provider-specific.
Our technology is informed by cutting-edge research from our Technology Advisor, Professor Haohan Wang. This research is from top-tier academic institutions and conferences.
Haibo Jin, Andy Zhou, Joe D. Menke, Haohan Wang
Jin et al. demonstrate that even sophisticated moderation guardrails can be bypassed using cipher character obfuscation, achieving 75% success rate against leading commercial and open-source models.
Haibo Jin, Ruoxi Chen, Peiyan Zhang, Andy Zhou, Yang Zhang, Haohan Wang
Jin et al. introduce a 4-role LLM system for automated security testing, achieving 82% average success rate in identifying guideline violations with natural-language attacks.
These papers form the foundation of GuardAI's understanding of attack vectors and defense mechanisms, enabling us to build protection systems that address real-world threats documented in peer-reviewed research.
Test real attacks against a GuardAI-protected system and see them get caught and rewritten. Or schedule a technical deep-dive with our team.