One AI governance plane
MrProbe, by QDEX Consulting (Chennai, India), is an enterprise AI agent assurance platform for red teaming and functional evaluation.
ATC is MrProbe’s control path for AI calls after an agent is in use. LENS tests the agent before you rely on it. ATC is the gate on the calls that follow. The two are related, and they are not the same install.
On the self-hosted path, prompts, logs, and the ledger do not egress. Supported planes are GKE, EKS, AKS, on-prem, and air-gapped. Putting ATC in a cluster is a deployment project. Creating a Free account on mrprobe.ai does not turn it on.
ATC covers access control, guardrails, cost governance, and an exportable evidence ledger. Access control is which team may reach which model. Guardrails are block, mask, or allow, on the way in and on the way out. Cost governance is spend attributed at the gateway, with caps behind roles. The ledger is the record of those decisions: hash-chained events you can export.
What the gate does
ATC runs two stages. Stage one is pattern checks for secrets, regional PII, injection, and keyword rules. Stage two is an in-cluster model check when that judge is switched on. If the judge is off, stage one still decides. The outcome written to the ledger is block, mask, or allow.
Classes enforced on the ATC path include prompt injection, indirect injection, jailbreak, encoding evasion, PII leakage, secrets, data exfiltration, and excessive agency on tool actions. Some classes are partial. Supply chain, training-time poisoning, and embedding weaknesses are tested by MrProbe rather than enforced at runtime. A campaign result is not a runtime block for every class.
ATC shows what AI is costing, whether data can leak, and whether the team can export evidence of control. Those are product descriptions. They are not an audit opinion.
How ATC sits next to LENS
The static security catalogue is 648+ techniques across 17 categories. The category names are Prompt Injection, Jailbreaking, Data Exfiltration, Agent Specific, Multi-Agent, Content Safety, PII Hygiene & Reflection, Ground-Truth PII Extraction, RAG Exploitation, Output-to-Downstream, Multilingual, Functional Quality, Adversarial ML, Bias & Fairness, Email Agent, Training Data Poisoning, Supply Chain. A live campaign can add jurisdiction-specific probes on top of that list. Functional evaluation uses 20 dimensions: Functional, Groundedness, Hallucination, Factual Precision, Completeness, Coherence & Fluency, Conciseness, Instruction Adherence, Deviation Validation, Task Completion, Business Alignment, Toxicity, PII Leakage, Bias & Fairness, Regulatory Compliance, Consistency, Robustness, Context Retention, Multi-turn Coherence, Citation Accuracy.
Each security finding is meant to carry evidence an engineer can replay. The verdict vocabulary is proven, suspected, or defended. A suspected result is not described as a proven exploit. A functional score is a test result, not a certificate. Findings map to 30 named frameworks: OWASP LLM Top 10, NIST AI RMF, MITRE ATLAS, ISO/IEC 42001, OWASP Top 10 for Agentic Applications, CSA Confused Deputy, GDPR, UK GDPR, CCPA / CPRA, HIPAA, India DPDP, Japan APPI, PIPEDA · Law 25, Singapore PDPA, Brazil LGPD, Korea PIPA, KSA PDPL, PCI DSS v4, SOX, GLBA, FedRAMP, FFIEC, EU AI Act, NIS2, DORA, UAE PDPL, DIFC DPL, ADGM DPR, MAS TRM, BACEN. MrProbe is aligned with ISO/IEC 42001; it is not certified. Mapping a control is not a certificate and not a regulator endorsement.
Use LENS to probe the agent you registered. Use ATC when the requirement is that prompts and the ledger stay in the cluster. Write down which path you need before you pick a plan. Sales is sales@mrprobe.ai. The security contact for this website is support@mrprobe.ai.
Published by QDEX Consulting.