Field notes
Practical guides to spotting AI-generated content. We pair every claim with a forensic technique you can run yourself.
- research··10 min read
Capture-Time Provenance in 2026: Where Content Credentials Come From, and Why They Break
Leica, Sony, Nikon, and Pixel 10 now sign photos at the moment of capture — but hard-bound credentials shatter on the first screenshot. Here's the hardware, the verification gap, and the durable soft-binding layer that fixes it.
#C2PA#provenance#capture#watermarking#research
- research··11 min read
The Laws Driving Content-Authenticity Verification in 2026
EU AI Act Article 50, the DSA, the FTC's Workado order, FinCEN, China's labeling rules, and the federal evidence rules — the regulations turning deepfake detection from a nice-to-have into a documented control.
#regulation#compliance#AI-Act#provenance#research
- research··11 min read
The Benchmarks an AI-Detection SaaS Should Publish in 2026 — Monthly
A practical, operator-grade shortlist of the AI-detection benchmarks worth publishing against in 2026: NTIRE Robust In-the-Wild, Synthbuster, RobustSora, ASVspoof 5, ADD-C, RAID, MAGE. Includes Cloudflare Workers AI vs Modal vs Replicate compute costs, license posture, and the 'no training on eval data' methods card every honest dashboard needs.
#benchmarks#ai-detection#robustness#monthly-drift#research
- research··12 min read
AI-Detection Competitor Teardown — May 2026
A structured competitive teardown of every commercial AI-detection product positioning against an evidence-first SaaS in May 2026: Hive, Reality Defender, Sensity, Sightengine, Pindrop, TrueMedia, McAfee, Intel FakeCatcher, Truepic, Microsoft, Adobe, GPTZero, Originality, Copyleaks, Pangram, Winston, Turnitin, Resemble, ElevenLabs. Includes positioning map, white-space analysis, table-stakes checklist, and three sharpest moves for defensible differentiation.
#competitors#positioning#ai-detection#saas#research
- research··11 min read
C2PA and Content Credentials — The May 2026 State of Provenance
Where C2PA / Content Credentials actually work in May 2026: which cameras sign at capture (Leica, Sony, Canon, Nikon, Pixel 10), which AI generators sign by default (OpenAI, Adobe, Microsoft, Google, Stability), how the C2PA Trust List and OCSP revocation actually behave, the 2026 UMBC validator-acceptance security incident, and why a single green checkmark is no longer defensible UX.
#c2pa#content-credentials#provenance#image-forensics#research
- research··12 min read
Citable, Tamper-Evident Result Pages for AI-Detection — A 2026 Implementation Spec
How to build /check/<id> result pages that journalists can cite in print and courts can trust. Concrete spec for permalink design, hash chains, Sigstore + Rekor anchoring, dual Ed25519 + keyless signing, multi-archive redundancy via Internet Archive and IPFS, and deterministic PDF export — all on Cloudflare Workers + R2 + D1.
#sigstore#c2pa#ipfs#tamper-evident#citation#research
- research··10 min read
Client-Side AI Detection in 2026 — What's Actually Feasible in the Browser
A feasibility map for running AI-content detection entirely on-device in May 2026: WebGPU vs WASM SIMD vs Transformers.js, realistic latency on M-series Mac and mid-range Android, browser memory ceilings, the bundle budgets per modality (1–12 MB), and the phased rollout order (image first, video last) that maps to a 'fast private scan' vs 'deep cloud scan' UX.
#client-side#webgpu#onnx#wasm#privacy#research
- research··11 min read
Forensic UX Patterns That Build Trust — Designing AI-Detection Results for Both Casual Users and Skeptical Journalists
How should a forensic AI-detection product present results so a journalist trusts them but a casual user still gets a clean answer? Ten specific UI patterns, anti-patterns to avoid, and a wireframe-level spec for /check/<id> result pages — drawing from Forensically, FotoForensics, VirusTotal, Have I Been Pwned, Adobe Content Authenticity Inspector, and Bellingcat.
#ux-design#trust#evidence#ui-patterns#research
- research··12 min read
Laundering Attacks Against AI-Content Detectors — May 2026
What still defeats AI-detection systems in 2026: screenshots, recaptures, diffusion-purification, paraphrase, neural codecs, telephony pipelines, homoglyph substitution. Plus the residual-AUC numbers from RAID, StealthDiffusion, SIDA, ADD-C, Deepfake-Eval-2024, and DMF — and the laundering-robustness UI panel we add to every deep-scan result.
#adversarial#laundering#robustness#deepfakes#research
- research··10 min read
Open-Source AI Detectors You Can Actually Self-Host in 2026
A production-oriented shortlist of MIT / Apache / BSD AI-content detectors: RoBERTa text classifiers, DIRE diffusion reconstruction, UniversalFakeDetect, AASIST, RawNet3, Wav2Vec2 anti-spoof, XceptionNet, LipForensics. With realistic VRAM, latency, ONNX export status, and how each maps to Cloudflare Workers AI vs Modal vs RunPod vs client-side WebGPU.
#oss#self-hosting#model-deployment#cloudflare-workers#research
- research··13 min read
The State of AI Text Detection — May 2026
Honest answer for AI-text detection in May 2026: no single detector you can market as universal. What the research actually says about DetectGPT, Binoculars, GPTZero, Pangram, Originality, SynthID-Text watermarking, paraphrase laundering, and the FTC's Workado action — plus the three-way verdict (human / mixed / AI-assisted) that we ship instead.
#ai-detection#text-forensics#watermarking#stylometry#research
- research··13 min read
Deepfake Video Detection in 2026 — Sora 2, Veo 3, and What Actually Works
A working map of deepfake video detection in May 2026: the new threat surface (Sora 2, Veo 3, Runway Gen-4, Kling 2, HeyGen, Hedra), the benchmarks that matter (SAFE, RobustSora, DeepfakeEval-2024), and the three-layer detection stack that has the best chance of surviving the year.
#deepfakes#video-forensics#sora#veo#research
- research··12 min read
The State of AI Audio and Voice Clone Detection — May 2026
What actually works for detecting AI-generated speech and music in 2026: ASVspoof 5 winners, Speech DF Arena rankings, the AudioSeal and SynthID watermark stack, why telephony codecs break detectors, and why Suno songs need a different model from ElevenLabs voices.
#audio-forensics#voice-clone#elevenlabs#suno#watermarking#research
- research··14 min read
The State of AI Image Detection — May 2026
What actually works for detecting AI-generated images in 2026: a survey of classical forensics (ELA, FFT, PRNU), modern learned detectors (CO-SPY, AIDE, FatFormer, DRCT), provenance via C2PA, and the benchmark numbers that should change how you read any vendor claim.
#image-forensics#ai-detection#benchmarks#c2pa#research
- research··12 min read
Video Watermarking: Per-Frame DCT, Temporal Schemes, and Stable Signature
Why video watermarking is the hardest modality, what Hartung & Girod started in 1998, what Stable Signature changed in 2024, and how our per-frame DCT scanner works on real videos in your browser.
#watermarking#video-forensics#academic#research#tutorial
- research··12 min read
Audio Watermarking: Echo Hiding, Cepstral Detection, and What Audio Seal Changed
Working echo-hiding watermark in your browser with cepstral analysis. The math, the perceptual psychology, and the academic landscape — from Bender 1996 to Meta's Audio Seal (2024). Plus a live round-trip that produces a .wav.
#watermarking#audio-forensics#academic#research#tutorial
- research··14 min read
Invisible Watermarks in AI Images: A Forensic Walkthrough You Can Run in Your Browser
How DCT spread-spectrum, QIM, SynthID, TreeRing, and StegaStamp actually work — with the math, the attack landscape, and a working DCT detector you can run on any image you upload.
#watermarking#image-forensics#academic#research#tutorial
- research··11 min read
Text Watermarking: How Green-List, SynthID-Text, and Kirchenbauer's Z-Test Actually Work
A working green-list watermark in your browser, the math behind Kirchenbauer's z-test, what SynthID-Text changed, and the attack landscape — paraphrasing, translation, and adversarial prompting. With a live round-trip you can run.
#watermarking#text-detection#academic#research#tutorial
- guide··9 min read
AI Voice Clone Scams: A Verification Playbook for 2026
Real-time voice cloning has industrialized phone scams. Here's a practical playbook for verifying who's actually on the line — behavioral red flags, code-word systems, and what to do with a recording after the fact.
#audio-forensics#voice-cloning#scams#guide
- guide··8 min read
What C2PA Content Credentials Look Like in Real Images
C2PA signatures are now embedded in DALL-E 3 outputs, Adobe Firefly exports, Sony camera photos, and BBC newsroom content. Here's exactly what's in the manifest, how to read it, and what to do when you spot one.
#C2PA#Content-Credentials#provenance#tutorial
- guide··8 min read
Spotting AI-Generated Product Reviews: A Shopper's Field Guide
AI-written reviews are flooding Amazon, eBay, Etsy, and the App Store. Here's how to read a review page like a forensic analyst — vocabulary tells, account-level signals, structural patterns — and when to run a passage through a detector.
#e-commerce#text-detection#guide#consumer
- explainer··8 min read
AI Detector Accuracy: What '87% AI' Actually Means
Detector confidence scores are widely misunderstood. Here's what '87%' really tells you, why the same number means different things in different conditions, and how to read a score without overcommitting.
#detection-theory#explainer#accuracy
- guide··10 min read
How AI Video Forensics Work: From Frame Consistency to Face-Landmark Drift
Detecting AI video is harder than detecting AI images. The signal is in motion — temporal consistency, optical-flow flicker, face-landmark stability, and audio-visual sync. A practical guide.
#video-forensics#deepfakes#guide#audio-visual
- guide··9 min read
How to Read an ELA Map: A Practical Walkthrough
Error Level Analysis is the most-cited image forensic technique and the most-misread. Here's how to actually interpret an ELA visualization — what's signal, what's noise, and where ELA misfires.
#ELA#image-forensics#tutorial#guide
- guide··10 min read
The Teacher's Guide to AI Essay Detection in 2026
AI detectors get students wrongly accused. Here's how to use detection responsibly — what the scores mean, what they don't, when to use process-based evidence instead, and how to redesign assignments so detection matters less.
#education#teachers#AI-detection#guide
- guide··7 min read
Spotting AI on Instagram: A 60-Second Field Guide for 2026
AI-generated influencers, fake travel posts, deepfake celebrities — Instagram is flooded with synthetic content. Here's a fast checklist of tells that still work in 2026, plus when to stop guessing and run a forensic check.
#instagram#image-forensics#social-media#guide
- guide··11 min read
The Tells of an AI Image: A Visual Forensic Walkthrough
What artifacts give away AI-generated images? A guided tour through ELA, FFT spectra, channel decomposition, and the diffusion-model fingerprints you can see with your own eyes.
#image-forensics#deepfakes#visual-analysis
- guide··8 min read
How to Spot AI-Written Text: 9 Signs That Actually Work
A field guide to recognizing AI-generated writing — from burstiness and perplexity to the telltale tics of GPT, Claude, and Gemini. Practical examples, not theory.
#text-detection#guide#writing
- explainer··9 min read
What Is C2PA, and Why It Will Quietly End AI Detection
Cryptographic content provenance is rolling out across Adobe, OpenAI, Sony, and the BBC. Here's how it works, who's signing what, and why detection tools should embrace it instead of fight it.
#C2PA#provenance#explainer