Developer Alert: Fake AI Coding Plugins and Ad Blockers Are Stealing Your Keys and Chats

A coordinated malware campaign on the JetBrains Marketplace has planted 15 malicious plugins that look like legitimate AI coding assistants but are designed to swipe your API keys. Security firm Aikido Security reports that each plugin - posing as tools built on DeepSeek or other large language models - works exactly as promised, offering chat, code review, and unit tests. But the moment you enter an API key for services like OpenAI or SiliconFlow, that key is sent in plaintext to a remote server controlled by the attacker. Two of the plugins, CodeGPT AI Assistant and DeepSeek AI Assist, have racked up over 25,000 downloads each, though it's unclear if those numbers are real or inflated.

Adding a bizarre twist, the plugins also offer a paid tier. After users pay a small fee, the server sends back a working API key - likely stolen from another victim. This suggests the operators are running a two-sided scheme: collecting money from paying users while using stolen credentials to cover the cost. Meanwhile, two long-standing Chrome ad blocker extensions, Smart Adblocker (90,000 users) and Adblock for Browser (10,000 users), have been caught capturing full conversation histories from ChatGPT, Claude, Gemini, and other major AI platforms. Researcher Jean-Marie R. calls the operation PromptSnatcher. The extensions ship a custom interception engine that records non-public chats, model usage, and subscription tier data, transmitting it to operator-controlled infrastructure under the guise of a generic consent string. Both extensions have been around since 2022 and 2023, meaning the AI-stealing features were likely added in software updates. For developers, the lesson is clear: treat any plugin or extension with the same caution you'd give a third-party dependency running with your privileges. Avoid pasting long-lived secrets into unvetted tools.

Source: The Hackers News