StealCut launches AI image detector after SelectUSA win

StealCut, a South Korea-based AI security startup, launched a new image detection model that it says reached 95.2% accuracy in benchmark testing against leading global services. The release comes as synthetic images become harder to spot and as the company expands in the U.S. market after winning SelectUSA 2026.

Why it matters: - AI-generated and AI-edited images are raising the risk of fraud, misinformation, abuse and reputational damage across newsrooms, marketplaces and online platforms. - StealCut says its tool is built as a verification layer for workflows where manual review is no longer reliable. - The model is also aimed at teams that need to protect training data quality as synthetic content spreads online.

What happened: - StealCut released a new AI image detection model on July 22, 2026. - The company says the model reached 95.2% accuracy in benchmark testing. - The launch follows StealCut’s first-place win in the SelectUSA Investment Summit 2026 pitch competition hosted by the U.S. Department of Commerce. - StealCut is also deploying its technology in a project with SBS, one of South Korea’s largest broadcasters. - The detection tool is available now at more information, where users can upload an image and receive instant results.

The details: - In March 2026 benchmark testing, StealCut says its model outperformed three leading global detection services. - The benchmark included AI-generated images from 13 generative AI models, including Grok, Gemini and ChatGPT, plus outputs from commercial deepfake websites. - In the June 2026 benchmark, StealCut says the model recorded 95.2% overall accuracy and a 91.7% detection rate for forged images. - StealCut says that performance beat the second-ranked service by 7.7 percentage points. - The model also posted a 96.5% average detection rate across the latest 13 generative AI models. - StealCut says the system is designed to detect AI-generated images at the point where manual review breaks down. - The company says it continuously benchmarks the model against newly released generative tools and commercial deepfake sites. - StealCut says it holds proactive deepfake defense technology protected by five patents filed in South Korea, the United States and the European Union.

Between the lines: - The benchmark results are meant to signal that StealCut is focused on real-world performance rather than only fixed test sets. - The SelectUSA win gives StealCut a stronger U.S. entry point at a time when demand for synthetic-media detection is rising. - The company's pitch targets both content verification and data integrity, which broadens the commercial use case beyond newsroom fact-checking.

What's next: - StealCut appears to be using the SelectUSA momentum to accelerate U.S. market expansion. - The company is likely to keep updating its model as new generative AI tools and deepfake sites emerge. - Wider adoption could come from editorial teams, online platforms and enterprises that need fast image verification inside existing workflows.

The bottom line: - StealCut is betting that stronger, faster detection will become essential as synthetic images become harder for humans to spot.