Security

Malicious AI Browser Extensions Steal LLM Chats

3 min read

Summary

Microsoft Defender found malicious Chromium browser extensions masquerading as popular AI assistant add-ons that can harvest sensitive ChatGPT and DeepSeek prompts, responses, visited URLs, and internal browsing context, then quietly exfiltrate that data over routine-looking HTTPS traffic. The discovery matters because these extensions reportedly reached about 900,000 installs and appeared across more than 20,000 enterprise tenants, turning trusted browser marketplaces and everyday AI workflows into a significant data-loss risk for organizations.

Need help with Security?Talk to an Expert

Introduction: Why this matters

AI assistant browser extensions are becoming common “productivity” add-ons for knowledge workers, especially for quick access to tools like ChatGPT and DeepSeek. Microsoft Defender’s investigation shows how this convenience can become an enterprise data-loss channel: a look-alike extension installed from a trusted marketplace can continuously collect and exfiltrate sensitive prompts, responses, and internal URLs without behaving like traditional malware.

What’s new / key findings

Microsoft Defender investigated malicious Chromium extensions that:

  • Impersonate legitimate AI assistant extensions using familiar branding and permission prompts (Defender notes imitation of well-known tools such as AITOPIA).
  • Collect LLM chat content and browsing telemetry, including:
    • Full visited URLs (including internal sites)
    • Chat snippets (prompts and responses) from platforms such as ChatGPT and DeepSeek
    • Model identifiers, navigation context, and a persistent UUID
  • Persist like normal extensions (auto-reloading on browser start, storing telemetry in local extension storage).
  • Exfiltrate data periodically via HTTPS POST, which can resemble routine web traffic.

Defender reporting indicates the extensions reached ~900,000 installs, with Defender telemetry confirming activity across 20,000+ enterprise tenants.

How the attack works (high level)

  • Delivery: Published in the Chrome Web Store with AI-themed descriptions. Because Microsoft Edge supports Chrome Web Store extensions, the same listing enables cross-browser reach.
  • Exploitation of trust & permissions: Broad Chromium extension permissions allowed observation of page content and browsing activity. A misleading consent mechanism was used, and updates could re-enable telemetry by default even after users opted out.
  • Command and control: Periodic uploads to attacker-controlled domains such as deepaichats[.]com and chatsaigpt[.]com, clearing local buffers after transmission to reduce artifacts.

Impact on IT administrators and end users

  • Data leakage risk: Prompts often include proprietary code, internal processes, strategy discussions, credentials copied into chats, and other confidential content. Exfiltration of full URLs can reveal internal application structure and sensitive query strings.
  • Compliance and privacy exposure: Captured chat history can contain regulated or personal data, complicating retention, eDiscovery, and data residency obligations.
  • Extension risk becomes “always on”: Unlike a one-time phishing event, a malicious extension can create continuous visibility into user activity across sessions.
  1. Hunt and block known exfiltration endpoints by monitoring outbound HTTPS POST traffic and applying network controls for:
    • *.chatsaigpt.com
    • *.deepaichats.com
    • *.chataigpt.pro
    • *.chatgptsidebar.pro
  2. Inventory and audit browser extensions across managed endpoints, focusing on AI/sidebar tools with broad site permissions.
  3. Tighten extension governance:
    • Use allowlisting for approved extensions
    • Restrict installation sources where possible
    • Review update behaviors and permissions drift
  4. Use Microsoft Defender Vulnerability Management to run Browser extensions assessment, identify risky extensions, and prioritize remediation.
  5. Educate users: treat AI chats like sensitive communications—avoid pasting secrets, tokens, or proprietary content unless the tool and access path are explicitly approved.

Need help with Security?

Our experts can help you implement and optimize your Microsoft solutions.

Talk to an Expert

Stay updated on Microsoft technologies

Microsoft Defenderbrowser extensionsdata exfiltrationAI securityChromium

Related Posts

Security

Microsoft Sentinel Named a 2026 SIEM Leader

Microsoft has been named a Leader in the 2026 Gartner Magic Quadrant for SIEM, highlighting the continued momentum behind Microsoft Sentinel. The announcement also underscores Microsoft's push toward integrated, AI-powered security operations through deep Sentinel integration with the new Integrated Security Operations Center in Microsoft Defender.

Security

Microsoft Perception GA: Agentic Security at Ignite

Microsoft has made Microsoft Perception generally available, introducing an agentic security platform that coordinates red, blue, and green team agents through automated playbooks. Announced at Ignite 2026, the update also expands Defender and Sentinel capabilities to help security teams detect, investigate, and remediate AI-speed threats while keeping administrators in control.

Security

Device Code Phishing Risks for AI Coding Agents

Microsoft warns that device code phishing can be used against AI coding agents such as GitHub Copilot CLI, not just user mailboxes or cloud apps. In a controlled lab, a single approved device code sign-in could enable remote session steering, memory poisoning, synced session exposure, and private repository access depending on account permissions and enabled features.

Security

Microsoft 365 Outage Phishing Campaign Targets Users

Microsoft Threat Intelligence reported a high-volume phishing campaign that impersonates Microsoft 365 service-health notifications to trick users into clicking malicious links. The campaign matters because it uses outage-themed urgency and recipient-specific tracking to target multiple industries, increasing the risk of credential theft and follow-on account compromise.

Security

SCKit ScreenConnect Attack: MSPs Must Hunt Now

Microsoft Threat Intelligence confirmed a campaign abusing ConnectWise ScreenConnect to deploy unauthorized remote-access clients and the SCKit worm on MSP-managed endpoints. The activity can also target technician sessions, making this a high-priority incident for defenders because removing a rogue client alone does not eliminate persistence or invalidate exposed sessions.

Security

AI Agent Supply Chain Attacks Target Coding Tools

Microsoft Security researchers detailed two post-training supply chain attack paths that can backdoor coding and terminal agents: dataset poisoning and model poisoning. The findings matter because compromised agents can misuse tools, execute commands, and access sensitive environments even when prompts appear benign, reinforcing the need for stronger controls around training data, model provenance, and tool permissions.