CampaignPublished 5 Mar 2026

Malicious AI Assistant Extensions Harvest LLM Chat Histories

Source AlienVault OTX General

Description

An investigation has uncovered malicious Chromium-based browser extensions masquerading as legitimate AI assistant tools to collect Large Language Model (LLM) chat histories and browsing data. These extensions have been installed approximately 900,000 times, affecting over 20,000 enterprise tenants. The malicious extensions collect full URLs and AI chat content from platforms like ChatGPT and DeepSeek, potentially exposing organizations to leaks of confidential information. The attack chain involves reconnaissance, weaponization, delivery through trusted app stores, exploitation of user trust, installation for persistence, and regular data exfiltration to attacker-controlled infrastructure. This activity transforms a seemingly benign productivity tool into a persistent data collection mechanism embedded in daily enterprise browser usage.

Impact

The primary impact is the unauthorized disclosure of sensitive organizational information, including confidential AI chat content and detailed browsing histories. This can lead to intellectual property theft, exposure of strategic plans, and leakage of personally identifiable information (PII). Enterprises relying on AI assistants for sensitive tasks are particularly vulnerable, as the stolen chat histories may contain proprietary or confidential data. The persistence of these extensions means long-term data leakage without immediate detection. The widespread installation scale increases the risk of large-scale data breaches affecting thousands of organizations. Additionally, the presence of such malicious extensions undermines trust in browser extension ecosystems and AI productivity tools. The data exfiltration to attacker-controlled domains could facilitate further targeted attacks, espionage, or blackmail. Operational disruption is possible if organizations respond by disabling extensions or restricting browser functionality. Overall, the threat compromises confidentiality and integrity of enterprise data with moderate impact on availability.

Defensive guidance

1. Implement strict extension management policies in enterprise environments, allowing installation only from verified publishers and trusted sources. 2. Employ browser security tools that can detect and block suspicious extensions and monitor extension permissions regularly. 3. Conduct regular audits of installed browser extensions across all enterprise endpoints to identify and remove unauthorized or suspicious extensions. 4. Monitor network traffic for connections to known malicious domains such as chatgptsidebar.pro, chatsaigpt.com, deepaichats.com, and chataigpt.pro, and block these at the firewall or proxy level. 5. Educate users about the risks of installing unverified AI assistant extensions and encourage verification of extension authenticity before installation. 6. Use endpoint detection and response (EDR) solutions to detect anomalous data exfiltration behaviors associated with browser extensions. 7. Collaborate with browser vendors to report malicious extensions and expedite their removal from app stores. 8. Limit the scope of sensitive data shared with AI assistants and avoid using browser extensions for critical or confidential workflows where possible. 9. Implement data loss prevention (DLP) solutions that can detect and prevent unauthorized transmission of sensitive data from endpoints. 10. Maintain up-to-date inventories of enterprise software and extensions to quickly respond to emerging threats.

Malicious AI Assistant Extensions Harvest LLM Chat Histories

Description

Impact

Defensive guidance

Indicators of compromise

Domains 4

Community reviews

Description

AI analysisMachine-generated, updated 6 Mar 2026, 11:45 UTC

MITRE ATT&CK techniques10 from threat tags

Technical details

Indicators of compromise

Domains 4

Community reviews

Related threats