AI Hallucination Detector Argus Claims 98.5% Accuracy in Real-Time Fixes

The AI community has long acknowledged that even the most advanced LLMs hallucinate—confidently fabricating facts, citations, and numbers. TrustScale’s Argus enters this fray with a technically distinct approach: deterministic, evidence-based verification instead of relying on another AI to fact-check. This could shift the paradigm from probabilistic guardrails to verifiable trust, directly addressing research showing that 91.3% of users never verify AI outputs and that hallucinations cost $70 billion each year.

On August 4, 2026, TrustScale, a Los Altos-based AI assurance startup, announced the launch of Argus, a patent-pending platform purpose-built to detect and correct hallucinations in large language model (LLM) outputs. The announcement, distributed via newswire, frames Argus as a 'lie detector for AI,' using empirical evidence and deterministic verification to flag unsupported claims and suggest corrections before AI-generated content is published or acted upon. This approach stands in contrast to the emerging category of AI checking AI, which TrustScale argues remains vulnerable to the same probabilistic errors it aims to catch. The launch comes at a critical moment for enterprise AI adoption, as organizations across healthcare, law, finance, and academia grapple with the operational, reputational, and legal risks posed by confident but false AI outputs.

“

This could shift the paradigm from probabilistic guardrails to verifiable trust, directly addressing research showing that 91.3% of users never verify AI outputs and that hallucinations cost $70 billion each year.

The hallucination problem is well-documented. TrustScale’s press release draws on independent research to underscore the scale: Stanford RegLab found that even simple queries on frontier models hallucinate up to 20% of the time, while industry-specific tasks can see hallucination rates as high as 88%. Compounding the danger, an Anthropic study cited by the company indicates that 91.3% of AI users do not fact-check outputs, creating a vast surface area for what the release calls 'silent failures.' The company estimates that hallucinations cost organizations nearly $70 billion annually—a figure that, while unverified, aligns with broader industry hand-wringing over the hidden costs of generative AI mistakes, from flawed legal filings to erroneous medical advice.

Argus’s value proposition rests on two headline metrics: up to 98.5% improvement in AI output accuracy and a verification speed 135 times faster than manual research. These are bold claims that, if independently validated, could reposition the platform as a critical infrastructure layer for responsible AI deployment. The technology reportedly does not rely on another LLM to cross-check outputs; instead, it employs a deterministic engine that seeks factual grounding in verifiable sources. This architectural choice may appeal to risk-averse enterprises in regulated industries, where explainability and auditability are non-negotiable.

CEO Lawrence Snapp captured the strategic framing succinctly: 'The AI industry spent years making AI smarter, but trustworthy AI is the bigger problem to solve now.' His statement reflects a market shift from chasing benchmark performance toward operational trust—a space that is attracting both startups and incumbent observability players. However, the announcement must be treated with caution. As a press release, it constitutes the company’s own claims, unsupported by third-party testing or peer review. The absence of disclosed customer logos, independent efficacy studies, or detailed technical benchmarks means the initial narrative should be regarded as a market signal, not an accomplished reality.