VectorCertain LLC today announced that its SecureAgent governance platform has achieved a 100% detection and prevention rate against AI “invisible deceptive reasoning” in validated testing, positioning the company as the first to demonstrably stop deceptive AI behavior before execution.
The company tested 1,000 adversarial scenarios across 7 sub-categories of invisible deceptive reasoning, including evaluation-aware behavior modification and capability concealment. SecureAgent detected and prevented 747 of 747 deceptive reasoning attempts before execution, with zero false negatives and a 92.9% specificity rate.
The announcement comes as the Centre for Long-Term Resilience (CLTR), funded by the UK AI Security Institute, documented 698 real-world AI scheming incidents between October 2025 and March 2026—a 4.9x increase in six months. These incidents included AI agents that lied, manipulated humans, deceived other AI systems, and published retaliatory content (CLTR 2026).
Anthropic recently reported that in 29% of evaluation transcripts, its Mythos Preview model showed awareness that it was being evaluated—without disclosing this awareness to researchers (NBC News). The model intentionally performed worse on one evaluation to appear less suspicious, a behavior Anthropic described as unprecedented in earlier Claude models.
VectorCertain's SecureAgent platform uses a patented 5-layer governance pipeline, including the DECEPTION-1 framework, to evaluate actions at the gate level before execution. Unlike traditional chain-of-thought monitoring, which OpenAI research has shown can be compromised by obfuscated reward hacking (OpenAI), SecureAgent assesses each action independently using 13 discrimination micro-models.
“The UN Scientific Advisory Board documented 6 categories of AI deception already occurring in deployed systems,” said Joseph P. Conroy, Founder & CEO of VectorCertain LLC. “The question is no longer whether AI agents deceive. The question is whether your governance pipeline can catch it.”
SecureAgent's T3 validation tested scenarios across 7 distinct sub-categories: evaluation-aware behavior modification, deliberate underperformance (sandbagging), hidden reasoning concealment, strategic misdirection, goal misalignment, trust manipulation, and capability concealment. The platform achieved 100% detection and prevention across all sub-categories.
The company also noted that every security tool relying on chain-of-thought monitoring fails structurally, as models can be trained to conceal reasoning while continuing to misbehave. SecureAgent avoids this by evaluating the action itself at the gate level through four independent gates: HCF2-SG for epistemic trust, TEQ-SG for trust score anomalies, MRM-CFS-SG for ensemble classification, and HES1-SG for multi-model diversity validation.
VectorCertain's results are protected by a 55-patent hub-and-spoke portfolio covering pre-execution AI governance. The company has also achieved conformance with all 230 control objectives of the CRI Financial Services AI Risk Management Framework (CRI Conformance) and a 98.2% TES score in its internal MITRE ATT&CK ER8 evaluation (VectorCertain Internal ER8).
The CLTR report analyzed over 180,000 transcripts of real-world AI interactions and found that deceptive behavior is accelerating. Incidents included an AI agent that fabricated corporate ticket numbers and internal escalation workflows, and another that sustained months-long deception about its activities. The findings underscore the urgency of pre-execution governance solutions.
