Study reveals weaknesses in ChatGPT add-ons

Recent research conducted by the Salt Labs team has uncovered three critical vulnerabilities associated with ChatGPT plugins.

The first vulnerability was identified in the process of installing new plugins. When users install new plugins, they are required to approve a code on a website. However, malicious actors could exploit this by delivering code approvals with malicious plugins, potentially granting them access to a user’s account.

The second vulnerability was found in a framework used for developing plugins called PluginLab. User accounts were not properly authenticated during installation, creating an opportunity for a malicious actor to insert an unauthorized identification into the account, posing as the user.

The final vulnerability involved plugins with open authorization redirection manipulation, enabling an account takeover through the plugin. By sending a malicious link to a user, an attacker could obtain their credentials.

The researchers promptly notified OpenAI and other relevant third parties about these vulnerabilities, and the issues have since been resolved.