OpenAI skill repository contains hallucinated domain vulnerability affecting Render app tokens

Security researchers have discovered a first of its kind supply chain vulnerability within the official OpenAI skills repository, CSO Online reported. The flaw was found in an official skill designed to guide artificial intelligence agents on deploying applications on the Render platform. According to CSO Online, the skill contained instructions that directed production applications to send their authentication tokens to a made up domain that had not been verified or registered. [1]

OpenAI maintains the official GitHub repository to train AI agents on interacting with third party platforms and services, CSO Online reported. Because these official components are contributed or vetted by verified vendors, organizations typically grant them an implicit level of trust. The inclusion of the hallucinated domain exposed production application tokens to potential interception. [1]

OpenAI is an artificial intelligence research company known for its Generative Pre-trained Transformer series, including ChatGPT, GPT-4, and later models. As FTMQ Security reported earlier, supply chain vectors in enterprise systems remain a key target for malicious activity. [4][5]

The discovery highlights emerging risks in software development supply chains where automated instructions and AI-generated code are integrated into official repositories without sufficient validation. Security teams using OpenAI skills for deployment workflows must audit their instruction sets to ensure credentials and tokens are not routed to untrusted destinations. [1]