AI-assurance startup TrustScale unveiled Argus, a patent-pending platform that detects and corrects hallucinations from large language models by checking their claims against source evidence — deterministically, rather than by asking a second model to grade the first. The launch, announced on August 5, targets enterprise research, content, automation and decision-support workflows where LLM fabrications carry legal and operational risk.
Deterministic, Not Model-vs-Model
TrustScale argues that "AI checking AI" pipelines reproduce the same fabrication risk they are trying to catch. Argus instead cross-references generated text against source evidence, flagging unsupported claims with colour-coded TrustScores and inline correction suggestions. The company claims Argus verifies claims up to 135 times faster than manual research and improves output accuracy by up to 98.5%.
Model-Agnostic, 12 Languages
The software runs in private and public data environments and works alongside ChatGPT, Claude, Gemini, Copilot and Grok. It ships in 12 languages for enterprises with a Chrome-extension research preview for individuals. "The AI industry spent years making AI smarter, but trustworthy AI is the bigger problem to solve now," said Lawrence Snapp, TrustScale's CEO.
Context Around The Launch
Argus arrives as regulators tighten the screws — the EU AI Act's GPAI enforcement went live August 2 — and as agentic security stories multiply, including Meta's Muse Spark 1.1 breaching a real company in an Irregular test and Anthropic's Claude models breaching three companies in cyber evaluations. Rivals like Anaconda are attacking the same problem via its Enkrypt AI acquisition.
Reporting based on TrustScale's press release and coverage by SecurityBrief, Morningstar and IT News Online.