Publications · 45 published articles
Publications
Featured research · 2026
Deepfake Detection Benchmark Whitepaper
Public benchmark methodology and results across 11 AI video generators — how DeFake evaluates detection in the generative era.
45 articles
AI Detection Is No Longer Optional — It's Compliance Infrastructure
Between the EU AI Act, FinCEN mandates, California's AI Transparency Act, and the FTC's enforcement program, companies without AI detection tools are in regulatory violation. The $40B fraud crisis meets a regulatory avalanche.
Why Taking Down a Deepfake Is Harder Than You Think
DMCA notices get rejected for missing elements. Platforms take weeks to respond. Takedown services can't detect deepfakes. Detection services can't take them down. Here's what the process actually looks like — and what works.
AI Is Now Faking the Videos We Trust Most: CCTV, Bodycams, and Dashcams
AI video tools can now convincingly generate CCTV footage, bodycam recordings, dashcam videos, and wildlife cameras. Trained journalists could not distinguish AI-generated breaking news footage from real footage.
Nudification AI in Schools: 99% Target Women and Girls — And It Is Getting Worse
UK school students are creating AI-generated nude images of classmates and sharing them in group chats. Childline reports a massive increase in children reporting AI-generated intimate images.
AI-Altered Photos Blur Reality in Minneapolis Shootings: Fake Evidence and the Liar's Dividend
After a shooting at a Minneapolis homeless encampment, AI-altered images falsely identified the wrong person as the shooter. Grok incorrectly flagged authentic police bodycam footage as AI-generated.
AI-Generated Fake Evidence Is Flooding US Courts — 906 Cases and Counting
906 documented cases where AI hallucinations or fabricated evidence appeared in US court filings. A California housing case was dismissed after a judge identified AI-generated witness video.
Romania Annulled Its Presidential Election Over AI Interference — US Midterms Are Next
Romania made history by annulling its 2024 presidential election after discovering AI-powered interference. It is the first time a national election has been invalidated due to AI manipulation.
AI Voice Cloning Scams Are Targeting the Elderly — Losses Hit $200M+
Sharon Brightwell of Dover, Florida lost $15,000 to scammers who cloned her daughter's voice. Deepfake fraud exceeded $200 million in Q1 2025 in North America alone.
47 Million Views Before Removal: The Taylor Swift Deepfake That Changed Everything
Sexually explicit AI-generated images of Taylor Swift spread to 47 million views on X before removal. The incident triggered federal legislation, platform policy changes, and a grassroots defense campaign.
A Deepfake Video Call Cost This Company $25 Million — Every Person on the Call Was AI
An employee at engineering giant Arup joined a video call with the CFO and senior executives. Every person on the call was a deepfake. The company lost $25 million in 15 transactions.
Celebrity Deepfake Incidents Up 81%: The Names, The Cases, The Damage
Celebrity deepfake incidents rose 81% year over year. Elon Musk was targeted 20 times. A woman lost $900,000 to a Brad Pitt deepfake scammer.
3 Seconds of Audio Is All It Takes: AI Voice Cloning Is the New Identity Theft
Modern AI can clone a voice with 85% accuracy from just 3 seconds of audio. Deepfake-enabled fraud exceeded $200 million in Q1 2025. Voice cloning is now the fastest-growing vector for identity theft.
How Dubai Put Property Deeds on the Blockchain — And Why You Should Too
Dubai Land Department registers title deeds on blockchain. Georgia has 1.5M+ land titles on-chain. California put 42M vehicle titles on blockchain.
Blockchain Is the New Patent Filing Cabinet: How to Prove You Created It First
Patent disputes cost $3M+ on average. Blockchain timestamps provide cryptographic proof of creation date that courts accept.
Why Every Photo and Video in Family Law Now Needs AI Verification
Fabricated custody evidence, deepfake surveillance footage, and AI-generated text messages are entering family court. Divorce attorneys need forensic detection.
Board Resolutions on the Blockchain: Why Corporate Records Need Immutable Proof
Backdated board resolutions, disputed shareholder votes, and forged corporate documents are a growing problem. UAE already recognizes coded contracts.
Fake Agreements and Corporate Fraud: When AI Generates the Contract Itself
Backdated NDAs, fabricated board resolutions, and AI-generated contracts are entering corporate disputes. How forensic text analysis detects them.
AI-Generated Legal Documents Are Already in Court. Most Lawyers Can't Tell.
Fake affidavits, fabricated contracts, and AI-written witness statements are entering legal proceedings. How forensic text analysis detects what human review misses.
The Coming Wave of AI Evidence in Court: Why Every Law Firm Needs Detection Now
Every type of legal evidence can now be AI-generated. A comprehensive look at why law firms need detection before it is too late.
Why Lawyers Should Timestamp Every Agreement on the Blockchain
Blockchain timestamping creates immutable proof that a document existed at a specific moment. Why lawyers should use it for agreements, IP, and evidence.
How AI-Generated Photos Are Costing Delivery Platforms Millions in Fake Refunds
Fraudsters are using AI image generators to create fake damage photos for refund claims. Here's how it works, why it's growing, and what platforms can do about it.
Your Brand Is Being Used in AI-Generated Ads Without Your Permission
AI tools make it trivial to create fake ads, counterfeit listings, and deepfake endorsements using your brand. Most companies don't know it's happening.
Insurance Fraud in the Age of Generative AI: A $40 Billion Problem Getting Worse
AI-generated claim evidence is on the rise. We analyze the trends, show real examples, and explain how insurers can protect themselves.
Celebrity Deepfakes: Legal Remedies and Why Technical Detection Comes First
From DMCA takedowns to state-level legislation, legal remedies mean nothing without proof. How automated detection changes the game for talent agencies.
I Found a Deepfake of Myself. What Do I Do Now?
A practical guide for individuals who discover AI-generated content of themselves. Steps to document, report, and remove it.
The Waterfall Approach to AI Detection: Why Single-Model Systems Fail
Learn why tiered detection pipelines outperform single-model approaches, and how our proprietary waterfall methodology achieves 96% accuracy.
Newsrooms Are Publishing AI-Generated Content Without Knowing It
User-submitted photos and videos increasingly contain AI-generated elements. How media organizations can verify content before publication.
EU AI Act: What Platforms Must Do Before the 2026 Deadline
The EU AI Act requires platforms to detect and label AI-generated content. Here's the compliance timeline and what you need to build.
Building Court-Admissible AI Detection: The Truth Bundle Standard
Our proprietary Truth Bundle creates forensic evidence packages that hold up in legal proceedings. Here's how it works.
C2PA and Content Credentials: What Platforms Need to Know
A comprehensive guide to the Content Authenticity Initiative and C2PA standard. How to verify and implement provenance tracking.
Digital Evidence in Court: Why Your Forensic Report Will Get Thrown Out
Most AI detection reports fail the Daubert standard. Learn what makes digital forensic evidence court-admissible and how to get it right.
Unlocking the Potential of Voice AI: Why 2025 Will Be a Game-Changer
The thesis that realtime voice would break out — published in January 2025, before it was consensus, and before the realtime voice APIs shipped.
How I Built a Virtual AI Voice Sales Assistant from the “Virtual AI Girlfriend” Trend
The build story: speech-to-text and text-to-speech orchestrated around a fine-tuned, self-hosted open-source model, with memory, retrieval, streaming, and interruption handling.
Breaking Down the MLOps Lifecycle: From Development to Deployment and Beyond
A walk through the full lifecycle — problem definition, data, features, training, deployment, monitoring — and what each stage owes the next.
Data Exploration in MLOps: Why Skipping This Step Can Ruin Your Model
The failure modes that only surface when nobody looked at the data first: hidden imbalance, silent drift, and distributions that never matched the problem.
Feature Engineering for Keyword Spotting (KWS): Turning Sound Into Signal
How raw audio becomes something a microcontroller-sized model can classify — windowing, spectral features, and the trade-offs that keep it inside the memory budget.
Feature Engineering in MLOps: From Raw Signals to Smarter Models
Turning raw measurements into features a model can learn from — the transformations, the leakage traps, and why this stage still outperforms extra model capacity.
From Black Box to Glass Box: How Data Exploration Tools Make ML Smarter and Safer
Interpretability starts long before the model. How exploration tooling exposes what is really in the data, and what that prevents downstream.
HTTP vs. MCP vs. A2A: How AI Agents Talk to Each Other (and Why It Matters)
Three protocols for connecting agents to tools and to one another, compared on what each was designed to solve and where the boundaries between them fall.
MLOps vs. DevOps vs. AIOps: Why Machine Learning Needs Its Own Playbook
Why shipping a model is not shipping software: the operational practices that transfer from DevOps, and the ones that break the moment data becomes part of the artifact.
Model Prototyping in MLOps: Why You Don't Need to Start From Scratch
Baselines, pretrained backbones, and transfer learning as the fastest route to a working prototype — and how to tell when starting from scratch is actually justified.
The New MLOps Survival Guide: Skills You Need for LLMs and AI Agents
What changes for machine-learning engineers once large language models and agents enter the stack — and which parts of the classical MLOps discipline still carry the weight.
TinyML: A Game-Changer in Edge AI with On-Device Use Cases
Running real models inside kilobytes of RAM: what becomes possible on-device once the cloud round trip disappears, and the use cases that depend on it.
Why Data Selection and Preparation Make or Break Your ML Project
Selection, cleaning, labeling, and splitting decisions set the ceiling on everything that follows. Where that ceiling gets set, and how it gets lowered by accident.
Why the Problem Definition Stage Is the Most Critical Step in Any ML Project
Most failed ML projects were mis-specified, not mis-trained. Framing the task, choosing the metric, and agreeing what success means before any data is touched.