OmniGPT Data Breach Exposes 34 Million Lines of User Conversations and Files
On February 10, 2025, a hacker leaked a substantial dataset from OmniGPT, a multi-model chat service. The exposed data includes emails, phone numbers, and tens of millions of chat lines, many of which contained sensitive credentials and business information that users had shared with the AI assistant. Researchers confirmed the authenticity of the leaked samples, highlighting how third-party AI aggregators accumulate secrets users inadvertently paste into chat interfaces.
Scope and Nature of the Exposed Data
The leaked dataset comprises approximately 34 million lines of user conversations, along with uploaded files. Among the exposed information are personal identifiers such as emails and phone numbers, which are typically used for account recovery or verification. More critically, the dataset includes credentials—such as passwords, API keys, and authentication tokens—that users may have shared during problem-solving or troubleshooting sessions.
The presence of business information suggests that corporate users, including employees or contractors, relied on OmniGPT for tasks requiring sensitive data handling. For example, developers might have pasted code snippets containing API keys, while executives could have shared financial or strategic details during discussions. The breach does not specify whether OmniGPT stored these interactions in plaintext or encrypted form, though the exposure of raw chat logs implies potential weaknesses in data retention policies.
Mechanism and Context of the Breach
While the exact method of the breach remains under investigation, the incident aligns with a broader trend of data exposure in AI-driven platforms. In OmniGPT’s case, the breach may stem from misconfigured access controls, insufficient encryption, or vulnerabilities in the platform’s infrastructure.
The leak also reflects a growing concern in the AI security community: the concentration of sensitive data in third-party chat logs. Users often treat AI assistants as trusted repositories for information they would not share elsewhere, assuming the interactions are ephemeral or secure. However, as demonstrated by this breach, such assumptions can lead to significant privacy and security risks when data is improperly stored or accessed by unauthorized parties.
Potential Impact on Users and Organizations
The exposure of credentials poses immediate risks for affected users, including credential stuffing attacks, unauthorized access to accounts, and potential financial or reputational harm. Industries such as finance, healthcare, and legal services—where sensitive data handling is critical—may face heightened exposure if employees continue to rely on third-party AI tools without proper safeguards.
Broader Implications for AI Security
OmniGPT’s breach is part of a larger pattern of data leaks in AI-driven ecosystems. Recent incidents, such as the exposure of 80,000 corporate AI logins through infostealer malware, highlight how stolen credentials can be weaponized to compromise AI accounts.
Organizations using AI tools must implement additional layers of security, such as data loss prevention (DLP) for sensitive inputs, encryption of chat logs, and regular audits of third-party AI integrations.
Investigative Status and Next Steps
As of February 10, 2025, the breach remains under investigation, with OmniGPT and relevant authorities assessing the extent of the exposure and potential mitigations. Users affected by the breach are advised to review their accounts for suspicious activity, enable multi-factor authentication, and rotate any credentials shared during interactions with OmniGPT. Organizations should conduct internal audits to identify unauthorized use of third-party AI tools and enforce policies that align with data protection regulations.
The incident serves as a cautionary example for both AI developers and end-users. While AI assistants offer convenience and efficiency, their adoption must be balanced with rigorous security measures to prevent unintended data exposure.