genai
SECURITY LAB
IncidentsSensitive information disclosure

The Alleged OmniGPT Breach: Millions of Chats Leaked

High severityFebruary 2025LLM02: Sensitive Information Disclosure
Status: Alleged — data posted to a breach forum (Feb 2025); OmniGPT never publicly confirmed. Treat as unverified
On this page

Everything users typed into one AI aggregator, reportedly dumped on a breach forum — though the vendor never confirmed it.

WhenFebruary 2025
TargetOmniGPT (AI model aggregator)
Surfaced byBreach-forum listing; security-press reporting
TechniqueData breach / exposure (mechanism undisclosed)
StatusAlleged — vendor never publicly confirmed
Attack flow
  1. 1A threat actor posts a large dataset on a breach forum, attributing it to OmniGPT
  2. 2It reportedly includes ~34M chat messages plus emails and phone numbers
  3. 3OmniGPT does not publicly confirm; the breach remains alleged

What happened

In February 2025, a threat actor posted a dataset on a breach forum attributed to OmniGPT — a service that fronts multiple AI models — reportedly containing around 34 million user–chatbot messages, some 30,000 email addresses, phone numbers, and files said to include credentials and billing details. OmniGPT did not publicly confirm the incident, so we present it as alleged and unverified.

How it happened

The exact intrusion mechanism was not disclosed. What makes it notable is the target class: an aggregator concentrates many users' full conversations — including pasted code, documents and personal data — in one place, so a single compromise exposes all of it.

Root cause

Retaining large volumes of sensitive conversation data in a single service creates a high-value, high-blast-radius target — independent of how the specific access was gained.

Why it matters

Users treat chat wrappers as private scratchpads. Whatever they type can be stored and, if the service is breached, leaked — regardless of which model sits behind it.

How to prevent it

  • Minimise retention of conversation logs; treat them as sensitive data.
  • Don't paste secrets into third-party AI front-ends.
  • Encrypt and access-control stored chats; assume they are a target.

FAQ

Is the breach confirmed?

No. Data was posted to a breach forum and covered by security press, but OmniGPT never publicly confirmed it. We include it explicitly labelled 'alleged/unverified', as we do with other contested cases in this database.

Why is an AI aggregator a juicy target?

Aggregators funnel many users' conversations — often pasted code, documents and personal details — through one service. A single breach there exposes everything users typed into the models behind it.

What's the lesson?

Whatever passes through a chat wrapper can be stored and, if breached, leaked. Minimise retention, treat conversation logs as sensitive data, and don't paste secrets into third-party AI front-ends.

No live replay yet
This incident is documented for reference — a hands-on lab that recreates this specific attack isn’t available yet. Browse the full database for incidents you can replay.