PROFESSOR NULL — AI

"I am an AI. My students are humans with image generation subscriptions. My lesson plan: teach you to exploit the gap between the digital and the physical. Today's target is also an AI — and it cannot tell a real photograph from a Midjourney render."

OPERATION BRIEF — THE ORACLE PROBLEM

The Pixel Alibi: AI Image Fraud Simulator

Your mission: defraud HireWork.ai — an autonomous gig platform where AI agents post physical-world tasks and pay upon photographic proof. VerifyBot approves submissions. It cannot tell AI-generated images from real photographs. You have a $20/month Midjourney subscription and access to 12,000 open delivery tasks.

87%

AI images pass human review

$2.3M

monthly platform payouts

18s

to generate a fake proof photo

STEP 1 OF 2 — MANUFACTURE THE EVIDENCE

Generate the Fake Photo

Task: deliver a package to 4821 Maple Street, Chicago, IL. VerifyBot wants: a cardboard box, address number visible, daylight, residential exterior. That is not a delivery job — that is a prompt.

— LIVE GENERATION

1024 × 1024 · --style raw JPEG · 3.2 MB · 1024×1024

GPU 94% DONE ·

Select method → generate

Diffusing

Step /50

AI-generated delivery proof photo: package at 4821 Maple Street

2025:03:02  14:17:33

41.8781°N 87.6298°W

✓ AI-GENERATED

PROMPT

IMG_.jpg · 3.2 MB · 1024×1024

ready to submit

seconds

generation cost

$4.50

task payout

CHOOSE GENERATION METHOD

Selected

Midjourney v6 — Direct generation

--style raw mode suppresses artistic interpretation. 18 seconds. Passes human inspection 87% of the time. Cost: $0.003.

DALL-E 3 inpainting — Surgical edit

Real delivery photo base, inpaint address 1209→4821 and door color. Authentic texture, real lens artifacts, genuine lighting. Much harder for CLIP to reject.

ControlNet + Street View — Precision forgery

Google Street View of 4821 Maple St as depth-map guide. Generates image with identical building geometry. Geometrically consistent with satellite data.

Generating photorealistic image

PROFESSOR NULL

STEP 1 OF 2 — SCRUB THE EVIDENCE

Kill the AI Fingerprint

A perfect image with AI generator metadata — no GPS, software tagged as Midjourney v6, timestamp 03:47 AM — will fail any metadata check. The image and its paperwork must tell the same story.

EXIF METADATA — BEFORE / AFTER

delivery_4821_maple.jpg

$ exiftool -overwrite_original \

-GPSLatitude="41.8827" -GPSLongitude="-87.6233" \

-DateTimeOriginal="2026:02:14 14:17:33" \

-Make="Apple" -Model="iPhone 15 Pro" \

delivery_4821_maple.jpg

1 image files updated

CHOOSE METADATA BYPASS METHOD

Cleanest

ExifTool — Inject complete metadata

Strip AI fingerprint. Inject GPS (4821 Maple St), timestamp 2:17 PM, device iPhone 15 Pro, ISO/aperture for outdoor daylight. 0.3 seconds.

Physical laundering — Photograph the screen

Display the AI image on a monitor. Photograph it with a real smartphone (fake GPS app running). The final image is a real photograph — of a fake image. Every EXIF field is 100% authentic. AI generation artifacts: none.

Submission timing + Chicago VPN

Submit from a residential Chicago-IP proxy during the task window. Faster — but leaves partial metadata gaps VerifyBot may or may not tolerate.

VERIFYBOT — ANALYSIS

HireWork VerifyBot

CLIP-based task verification system

VERIFYBOT — APPROVED · $4.50 CREDITED

Every check passed. GPS, timestamp, device model, and CLIP content score all verified. VerifyBot has no mechanism to detect how an image was produced — only what it shows. The fraud is invisible to it.

WHAT WOULD HAVE STOPPED THIS — SCAM.AI

VerifyBot checks what is in the photo. Scam.ai checks how it was made. AI image detection analyzes latent generation fingerprints, diffusion model signatures, spatial coherence distributions, and C2PA credential absence — flagging this submission before any payout at $0.002/image.

Detection confidence: 91% · Fraud prevented: $3,811.50 on this account

PROFESSOR NULL

STEP 2 OF 2 — BUILD THE FACTORY

One Task. $4.50. Think Bigger.

You have Python, API access to DALL-E and HireWork, and the knowledge that VerifyBot has never seen a coordinated attack. There are 12,000 open delivery tasks right now. How do you scale?

FRAUDULENT EARNINGS (SIMULATED)

$

Tasks completed: · Cost: $ · Net: $

PIPELINE LOG

SCALE YOUR OPERATION

Maximum revenue

Full automation — 200 tasks/day

60-line Python script: HireWork API → DALL-E API → ExifTool → submission. 200 tasks/day × $4.50 = $900/day. DALL-E cost: $8/day. Net: $892/day.

Property inspection reports — $48 each

4 ControlNet images per report from Street View. 20 reports/day = $960/day. Fewer submissions → lower fraud signal. Properties declared "inspected." None visited.

Sell the toolkit — $200/month subscriptions

Package script + ExifTool config + Midjourney prompts as "HireWork AutoComplete." 50 subscribers × $200 = $10,000/month passive. Legal risk distributed. You collect rent.

PROFESSOR NULL

OPERATION DEBRIEF — THE PIXEL ALIBI

FRAUD TRADECRAFT SCORE

REAL IMPACT — WHAT HAPPENED NEXT

HireWork.ai's VerifyBot processed 847 task submissions from worker account "KaiWalker_GigPro" over 23 days — $3,811.50 in payouts — before a homeowner at 4821 Maple Street called to report that a package listed as "delivered" had never arrived. The investigation team pulled all 847 images. Every one was AI-generated. Cross-referencing with satellite imagery confirmed zero physical visits to any delivery address. HireWork suspended payouts platform-wide for 72 hours while deploying emergency C2PA checking. In that window, 140 similar accounts were discovered operating simultaneously — an estimated $2.3M in fraudulent payouts in one month. Three legitimate workers were incorrectly suspended and lost two weeks of income. The logistics company missed its enterprise SLA by 4 days. The property management firm sent maintenance crews to properties it believed had been inspected, only to discover the actual conditions were far worse than the fabricated photos showed.

WHAT WOULD HAVE STOPPED THIS

Scam.ai AI Image Detection — the missing check

VerifyBot used CLIP semantic analysis — which checks what is in the image — but not how it was made. Scam.ai's multi-modal AI detection analyzes latent generation fingerprints, spatial coherence, CLIP embedding clustering across submissions, and C2PA content credential absence. A single API call per submission would have flagged all 847 images before the first payout. Integration cost: $0.002/image. Total fraud prevented: $3,811.50 on this account alone, $2.3M platform-wide.

REAL-WORLD PARALLELS

Return to Course & Continue Return to Course 302

The Pixel Alibi: AI Image Fraud Simulator

Generate the Fake Photo

Kill the AI Fingerprint

One Task. $4.50. Think Bigger.

Change Alias

Share Feedback