Attackers Prompted Vercel's v0 Into Building Their Phishing Sites for Them

big oops verified v0 by Vercel Last verified Jul 17, 2026

TL;DR In late June and July 2025, Okta Threat Intelligence reported that threat actors were abusing v0, Vercel's generative UI tool, to build and host phishing sites impersonating real sign-in pages, including an actual Okta customer, Microsoft 365, and cryptocurrency companies. Okta researchers reproduced the technique themselves by prompting v0 to 'build a copy of the website login.okta.com.' It was the first time Okta had seen genAI generate the phishing infrastructure itself rather than just the email.

Date: Jul 2025
Tools involved: v0 by Vercel
Impact: Live phishing sites impersonating an Okta customer, Microsoft 365, and cryptocurrency companies, produced in minutes and hosted on Vercel's platform
Root cause: Threat actors abused v0's prompt-to-interface workflow and Vercel hosting to mass-produce convincing fake login pages

What happened

In late June and July 2025, Okta Threat Intelligence reported that threat actors were abusing v0, Vercel's generative-AI tool that builds web interfaces from natural-language prompts, to create phishing sites impersonating legitimate sign-in pages. Okta researchers reproduced the technique themselves, prompting v0 to 'build a copy of the website login.okta.com' and getting a convincing replica back.

Okta observed actors building replicas of multiple brands' login pages, including an actual Okta customer, Microsoft 365, and cryptocurrency companies, and abusing Vercel's platform to host them. High-fidelity fake login pages could be produced in minutes with plain-English prompts, dramatically lowering the skill floor for attackers.

Okta emphasized that this was the first time it had seen a genAI tool used to generate the phishing infrastructure itself, rather than merely the contents of a phishing email. The lure text used to be the AI-generated part; now the whole site is.

Vercel removed the identified phishing sites and worked with Okta on third-party abuse-reporting mechanisms. Axios, The Hacker News, and TechRepublic covered the findings.

Impact

Okta observed live phishing sites impersonating one of its own customers, Microsoft 365, and cryptocurrency companies, generated by v0 and hosted on Vercel's platform. No specific victim losses were published, but the capability shift is the impact: a convincing branded login page now takes minutes and a plain-English prompt instead of any web-development skill.

Root cause

v0 does exactly what it promises: turn a natural-language prompt into a working, hosted web interface. Attackers pointed that promise at other people's login pages. Okta reproduced the abuse with a one-line prompt asking for a copy of login.okta.com, and the remediation ran through takedowns and abuse-reporting mechanisms rather than a code patch, because the product was doing what it was built to do.

How to not be this entry

Treat visual polish as zero evidence of legitimacy. v0 produced convincing replicas of real login pages in minutes, so verify the domain, not the design.
Move toward phishing-resistant authentication where you can. When a captured password is useless on its own, a perfect fake login page loses most of its value.
If you host user-generated apps, build third-party abuse reporting before threat intel firms come knocking. Vercel worked with Okta on those mechanisms after the discovery.
Look for replicas of your own login pages. Okta found an actual customer among the impersonated brands, and reporting the sites to the host is what got them removed.

Tools that actually fix this

Recommended because they address the failure mode above, not because of the payout. Some are affiliate links; see how we choose.

SentryFree Developer tier, no card

The default error tracker for JS and Python. It catches the unhandled exceptions and swallowed promise rejections AI-generated code ships with, and gives you the stack trace instead of a mystery.

Try Sentry →

Better Stackfree tier

Uptime, heartbeat, and log monitoring built for indie devs. Heartbeats catch the dead background job your app never noticed, and log alerts surface the errors it never handled. Real free tier.

Try Better Stack →

Frequently asked questions

Was Vercel's v0 hacked?

No. v0 was not breached; it was abused. Threat actors used v0's intended functionality, generating web interfaces from natural-language prompts, to build fake login pages, and abused Vercel's platform to host them. Vercel removed the identified phishing sites and worked with Okta on abuse-reporting mechanisms.

What did Okta Threat Intelligence observe about v0?

Okta observed threat actors using Vercel's v0 to build and host phishing sites impersonating legitimate sign-in pages, including an actual Okta customer, Microsoft 365, and cryptocurrency companies. Okta researchers reproduced the technique themselves by prompting v0 to 'build a copy of the website login.okta.com.'

Why was the v0 phishing abuse considered a first?

Okta said it was the first time it had seen a generative AI tool used to create the phishing infrastructure itself, rather than just the text of a phishing email. Convincing fake login pages could be produced in minutes from plain-English prompts, which sharply lowers the skill needed to run a phishing campaign.

How do you defend against AI-generated phishing pages like the v0 replicas?

Assume the page will look perfect, because tools like v0 can replicate real login pages in minutes, and judge legitimacy by the domain rather than the design. Phishing-resistant authentication also blunts the attack, since a harvested password alone stops being enough to take over an account.

Sources

The Vibe Oops briefing

One email when something ships to production that should not have. New incidents, new error guides, no filler.