Congratulations! You've won 7 nights at a beach resort
From: Rewards Desk
AI-generated01 · Prize and holiday scams
A dream resort, a deadline, and a small "booking fee". The resort photo is the hook, and it was generated.
Phishing emails increasingly use AI-generated photos, headshots and documents to look convincing. Expose them before you act. Save an image from a suspicious message, drop it into the dashboard or send it to the REST API, and see in seconds whether it looks AI-generated.
Your new home on Maple Grove Drive
James Miller <[email protected]>
to me · 9:41 AM
Hi Sarah, the house is still available. Here's a photo from the front. Send the $1,500 holding deposit today and it's yours.

Scammers use generated photos because they look real and can't be found with a reverse image search. A Chrome extension that checks images where you read your email is coming soon, and the steps below show how it will work. Today, check an image from an email in the dashboard or through the REST API.


2 images found
Step 1
The extension runs in Chrome next to your webmail. When you open a message, it finds the images in it on its own: no saving, uploading or forwarding by hand.
SynthID watermark
Google Gemini · OpenAI GPT-Image
Content credentials
C2PA manifest · EXIF metadata
Synthetic media model
Pixel-level forensic analysis
Step 2
The same checks as the Sightova dashboard: hidden AI watermarks, signed content credentials, and our own model reading the pixels for signs of generation.
AI-generated
No AI detectedStep 3
Each image gets a red or green label right where you see it, along with the reason: which watermark was found, or how confident the model is.
The emails are made up, but the images are not: two were generated with Google Gemini and one is a real phone photo. Each label below is what our checks returned for that file.
Congratulations! You've won 7 nights at a beach resort
From: Rewards Desk
AI-generated01 · Prize and holiday scams
A dream resort, a deadline, and a small "booking fee". The resort photo is the hook, and it was generated.
Members only: 70% off designer leather bags
From: Atelier Outlet
AI-generated02 · Fake shops and offers
Too-good-to-be-true deals need product shots. Generated ones cost nothing and show stock that doesn't exist.
Receipt from Tuesday's lunch
From: A colleague
No AI detected03 · Real photos stay green
Most images in your inbox are genuine. This phone photo of a receipt raised no signals, so it stays green.
Results are from Sightova's production checks on the original files. The AI images were generated by Sightova with Google's Nano Banana Pro and Nano Banana 2. Receipt photo from CORD v2 (NAVER Clova AI), licensed CC BY 4.0, cropped. Email subjects, senders and the mailbox are illustrations.
Spot a fake sender. Check whether a headshot or profile photo looks AI-generated, and get an AI probability and a verdict (ai-generated, real or uncertain) in seconds.
Know when an image was made by Gemini or GPT-Image, because the watermark says so. Each check looks for invisible Google SynthID and OpenAI watermarks that can name the exact tool.
Reads C2PA content credentials and EXIF metadata to reveal where an image came from. The model reads the pixels too, so the check works even when metadata has been stripped.
See exactly where an image looks AI-edited. The heatmap shows reviewers where to look. In the dashboard on the Starter plan and up.
Expose edits in an image of an invoice, receipt or similar document. The document tampering endpoint (Starter plan and up) flags regions that look edited, before anyone pays.
Drop an image from a suspicious email into the dashboard, or connect through our REST API from your own tooling, up and running with an API key. Start free with no credit card.
Traditional email security filters text, domains, and sender reputation — but phishing increasingly leans on the images in a message. AI-generated headshots pass DMARC checks. Forged invoices evade keyword filters. Sightova checks what those filters skip: whether an image looks AI-generated or edited.
Email remains the primary attack vector for cybercriminals, but the nature of phishing has fundamentally changed. Modern business email compromise campaigns no longer rely solely on spoofed domains and urgent language. Attackers now embed AI generated headshots to fabricate convincing sender identities, forge corporate letterheads pixel by pixel, and attach manipulated invoices that pass cursory human review. Organizations that rely exclusively on DMARC, SPF, and keyword scanning are missing a threat surface that an ai image detector exposes.
The most effective phishing emails in 2025 look nothing like the "Nigerian prince" spam of a decade ago. They arrive with a photorealistic AI generated headshot of a fictional CFO, a pixel perfect company letterhead, and an attached invoice that matches the target organization's actual vendor formatting. Every visual element is specifically designed to exploit the trust gap that text filters leave wide open.
Generative AI already enables attackers to produce headshots that pass reverse image searches, create branded PDF attachments with perfect font matching, and synthesize voice messages that complement the visual deception. The same open source model proliferation fueling deepfake generation is also powering a new generation of phishing toolkits that package sender photo, letterhead, invoice, and follow up voice note into a single automated pipeline.
The FBI reports that business email compromise caused over $2.7 billion in losses in a single year, and that figure continues to climb as AI lowers the barrier to entry for sophisticated fraud. Industries with complex supply chains, frequent vendor payments, and distributed workforces are especially vulnerable. By 2030, researchers predict that AI generated phishing content will outpace manually crafted attacks by a factor of ten.
The next generation of attack toolkits will offer fully automated email package creation, tailored to a specific target using scraped LinkedIn data and corporate filings. One text prompt will generate the sender persona, the letterhead, the invoice, and even the follow up correspondence. Defending against this shift calls for the same multi layered approach that modern cybersecurity AI detection brings to network security, including a check of the images in a message.
The target is not just large enterprises. Small and midsize businesses with lean finance teams and limited security infrastructure are increasingly in the crosshairs. A single successful BEC attack can represent months of revenue for a 50 person company, making the return on investment for attackers exceptionally high.
When you submit an image from a suspicious email, Sightova returns an AI probability and a verdict, looks for invisible Google SynthID and OpenAI watermarks, and reads C2PA content credentials and EXIF metadata. A watermark or credentials can name the exact tool behind an image, and the model reads the pixels, so it works even when metadata has been stripped. A flagged headshot gives you a clear reason to look closer at the sender.
For an image of an invoice or a similar document, the document tampering endpoint (Starter plan and up) flags regions that look edited and shows a heatmap. We test on 250k+ benchmark images and 50+ generative models and publish the results in our testing methodology.
Check an image in the dashboard, or connect through our REST API from your own tooling. For financial institutions already managing payment fraud through banking fraud detection systems, image checks add evidence for reviewers. Sightova is hosted in Germany, images are deleted after analysis and not used for training, and you decide what happens next.
The FBI reports that business email compromise caused over $2.7 billion in losses in a single year. Before you act on a suspicious message, check its images with Sightova.