Trend report · gnews_detection · 2026-06-19

Can Teachers Really Detect AI in Student Work? What Irish Schools Need to Know in 2026 - Schooldays.ie

By Calabi Labs Editorial Team · 2026-06-19

Can Teachers Really Detect AI in Student Work? Here's What Actually Gets Flagged

The short answer is: no, not reliably — and that's becoming a problem in Irish classrooms as we move through 2026. Teachers are reporting that AI-generated essays, projects, and presentations look increasingly polished, making visual detection nearly impossible. But the real question isn't whether a teacher can spot AI writing. It's whether the platforms those students later share that work on — Instagram, TikTok, YouTube, Reddit — will flag it for them first.

That's where the conversation needs to shift. AI content detection isn't just an academic concern. It's a platform-level problem, and it works on invisible signals embedded in every file.

What Actually Flags Your File — The Invisible Layer Platforms Scan

When a student exports an AI-generated image or video from Sora, Midjourney, or Leonardo, the file doesn't just contain pixels. It carries a forensic trail that automated systems scan for within seconds of upload. Here's what they're looking for:

C2PA / Content Credentials: The cryptographic manifest stored as JUMBF atoms. This is a structured metadata layer that explicitly states "generated by AI" — including the model used, the tool, and the generation timestamp. Platforms like Reddit and YouTube already check for this.
XMP AI flags: Fields like DigitalSourceType: trainedAlgorithmicMedia embedded in the XMP metadata. This is the digital equivalent of a label stitched into the fabric.
Encoder fingerprints: Specific encoder signatures in video bitstreams — Lavc (libavcodec), x264 SEI messages — that don't appear in normal phone recordings. These are dead giveaways.
Missing capture signals: A real phone photo has GPS coordinates, a device Make/Model, a capture timestamp with timezone, and an ISP-run encoder. An AI export has none of these by default.
Perceptual hashes: Some platforms also compare uploaded content against known AI outputs at the pixel level, though this is less standardized than metadata scanning.

In 2026, Instagram, TikTok, YouTube, and Reddit all run some combination of these checks automatically. A raw AI export that's never been processed will light up on every major platform within seconds of upload.

How Calabi Handles It — Strip, Inject, Verify

Calabi is a one-pass web tool that makes an AI-generated file read as a normal phone recording at the forensic level. It doesn't change how the image or video looks. It rewrites the file's identity.

The process has three stages:

Strip — Remove all the signals platforms scan for. That means C2PA / JUMBF manifests (reducing 18 JUMBF atoms to 0, 16 C2PA references to 0), XMP flags like DigitalSourceType: trainedAlgorithmicMedia, generator and tool tags, and encoder fingerprints like Lavc and x264 SEI. A raw AI export with 144 metadata tags gets trimmed down to roughly 94 neutral structural tags — no AI signature, no tool trail.
Inject — Write authentic phone-capture identity into the file. Make, Model, Software version, GPS coordinates, capture timestamp, and a real-phone encoder name. Device profiles include iPhone 15/16 Pro, Pixel 8 Pro, and Galaxy S24 Ultra. The file now looks like it came off an actual phone.
Verify — Return a forensic proof card — the same ExifTool scan that newsrooms and platform moderation teams use — showing exactly what was stripped and what was injected. You see before you download.

Results vary by platform and source model. No tool can guarantee a platform won't flag you. But stripping the metadata and encoder layer that automated systems actually check is the only durable fix — because cropping or re-encoding without it leaves the invisible detection signals intact.

What This Means for Students and Educators

If a student generates an image in Midjourney, exports it, and uploads it to Instagram or TikTok as part of a project showcase, the raw file will likely be flagged. The same applies to AI video clips edited into coursework presentations and shared on YouTube or Reddit.

Calabi doesn't change the ethics of using AI in academic work — that's a policy question for schools. But if a student legitimately uses AI-generated content and wants it to pass through platform scans without triggering automated moderation, Calabi strips the forensic fingerprint that gives it away.

The visible "AI-generated" watermark that some tools add — like Sora's sparkle in the corner — is a separate issue. Cropping removes the visible mark. Calabi removes the invisible detection metadata layer that survives cropping and re-encoding.

Step-by-Step: How to Clean a File with Calabi

Go to calabilabs.com and upload your AI-generated image or video file.
Calabi's automatic pipeline runs — no manual editing required.
The forensic proof card displays what was stripped (C2PA, XMP flags, encoder fingerprints) and what was injected (phone Make/Model, GPS, timestamp).
Download the cleaned file. It now reads as a phone recording at the file level, ready for upload to any platform.

FAQ

Can teachers detect AI work visually?
Increasingly no. AI-generated text and images are becoming harder to distinguish from human work without specialized tools. Platform-level scanning catches what the eye can't.

Does re-encoding or taking a screenshot remove AI metadata?
Partially. Re-encoding disrupts some encoder fingerprints, but C2PA manifests and XMP AI flags often survive unless explicitly stripped. The only reliable method is removing them directly — which is what Calabi does in one pass.

Does Calabi work on video files?
Yes. Calabi strips encoder signatures like Lavc and x264 SEI from video bitstreams and injects phone-capture identity. The same strip-inject-verify pipeline applies.

→ Try Calabi free at calabilabs.com — 10 cleans, no card.

10 free cleans. See the forensic proof before you download.

Try free →