01 / A MOVING TARGET
AI creates.
AI tries to catch up.
New forgeries prompt better detectors. New techniques then challenge those improvements. Relying on after-the-fact detection means facing a continuing cycle of adaptation.
Next-generation software
for media authentication.
KIMENWA is developing software to help authenticate captured photos and videos. From insurance claim submissions and site inspections to payments and public information, we’re preparing a new basis for trusting the media behind important decisions.
WHY IT MATTERS
Why does media authentication matter?
Real decisions.
Uncertain evidence.
One altered image can change a payment, an inspection, or an entire story.
Explore the moments when knowing what was actually captured matters.
AI-generated original and altered scenariosAll amounts are illustrative estimates in USD
THE CHALLENGES
Why is a new approach needed?
As AI forgeries evolve,
detection has to keep up.
Analyzing photos and videos for signs of manipulation is one approach to media verification. But as new generation and editing techniques emerge, the systems built to recognize them must keep evolving, too.
MEDIA ARCHIVE
SHEET 025 / 120REMAINING 2,874
AI creates.
02 / A GROWING WORKLOAD
One suspicious image.
A whole queue to review.
When an archive is screened without knowing where a forgery is, authentic media enters the queue, too. More material means more processing, and updated detection criteria may call for another review.
03 / PERFORMANCE IN CONTEXT
A different test can
tell a different story.
61–70%Accuracy reported in cross-generator evaluation
A 2023 study cited by NIST evaluated images from generators different from those used in training. This is a research result, not the current detection rate of commercial services. NIST AI 100-4, §3.2.2.1 ↗
04 / PROVENANCE & WATERMARKS
Records and signals
also have limits.
Provenance helps verify whether a record has changed, but does not guarantee that the scene is true. Watermark resilience also depends on the method and transformations; removal and spoofing remain risks.
AI detection, provenance and watermarks each contribute useful evidence. No single approach guarantees the authenticity of every piece of media.
Sources and evaluation context
The figure above is accuracy from a specific study cited by NIST, not a comparison between a company’s advertised and observed performance. The animation illustrates a full review of 3,600 media items; it does not represent measured throughput or a universal requirement to scan every file.
OUR APPROACH
What is KIMENWA working to provide?
A new paradigm
for media authentication.
KIMENWA aims to develop a new paradigm that addresses the limitations of existing media authentication methods. Starting with the challenges that remain in detection, provenance and watermarking, we’re rethinking the foundations of how media can be authenticated. Our focus is on evidence of a real capture, helping people make informed decisions about the photos and videos they receive.
A shift in the question we ask
The question behind detection
Are there signs
of manipulation?
Analyze the finished photo or video for clues that it has been altered.
The question we’re exploring
Is this a record
of a real capture?
Focus on authentication evidence that can support an answer.
PROJECT UNVEIL
When will the technology be revealed?
KIMENWA’s core technology
is patent pending.
KIMENWA has filed 38 patent applications in South Korea. Technical details remain confidential; we plan to introduce the technology behind this approach at the project unveiling.
Patent applications pending · Details under wraps
Project unveiling · November 23, 2026 KST
—Days
—Hours
—Minutes
—Seconds
Built with
From an idea
to what comes next.
KIMENWA is taking shape with ChatGPT and Codex, supporting research, design, and development as we prepare the project for its reveal.
ChatGPT
Codex