Smishing Attack on OpenAI Vendor Compromised Thousands of API Accounts
Smishing Attack on OpenAI Vendor Compromised Thousands of API Accounts#### One fraudulent SMS was enough for attackers to gain access to the analytical data of thousands of OpenAI API accounts. Not through a direct attack on OpenAI, but through their supplier Mixpanel.
In the new edition of Security Sunday, we will look at How smishing compromised Mixpanel and what it meant for OpenAI. The Shai-Hulud 2.0 worm, which spreads via npm packages and steals developer tokens. Australian man sentenced to 7+ years for evil twin Wi-Fi attacks at airports and GreyNoise IP Check: find out if your network is working for hackers.
Smishing Attack on OpenAI Vendor Compromised Thousands of API Accounts
OpenAI announced a security incident in which an attacker did not breach OpenAI’s systems directly, but exploited a vulnerability at external vendor Mixpanel, obtaining a limited set of analytics data about some OpenAI API users. This was not a leak of chats, prompts, API keys, or payment information, but rather a classic supply-chain problem where a third party proved to be the weakest link.
Mixpanel, a popular product analytics platform, was targeted by a smishing attack on November 8–9, 2025. This type of attack uses fraudulent SMS messages to trick employees into revealing credentials or clicking malicious links. The attacker gained access to part of Mixpanel’s environment and exported a dataset containing identification and analytics information about some customers, including OpenAI, which used Mixpanel to measure interactions on platform[.]openai[.]com for API users.
From OpenAI’s perspective, only metadata associated with API accounts was involved, not content. Specifically, this included the name and email associated with the API account, approximate location derived from browser or IP (city, region/state, country), operating system and browser information, and internal organization or user IDs. No passwords, API keys, chats, API request logs, usage data, or payment information were leaked.
The incident affected only some API users, not regular ChatGPT web or app users, who according to available information are not included in the leaked dataset.
After receiving details, OpenAI began directly informing affected organizations and users and completely discontinued using Mixpanel in production. Mixpanel secured the compromised accounts, terminated sessions, reset employee passwords, blocked attacker IP addresses, and claims all directly affected customers were notified. If you have not received an email from OpenAI, you should not be among those affected.
When an npm Package Becomes a Weapon
Let’s stay with supply-chain attacks, specifically focusing on the popular Zapier platform. Attackers took control of Zapier’s npm account, which manages packages like zapier-platform-core and zapier-platform-cli. They then infected these packages with the malicious Shai-Hulud 2.0 worm, which steals sensitive information during installation and spreads across the npm ecosystem.
Zapier quickly responded by pulling the compromised versions. The worm was inserted in so-called pre-install or post-install scripts, meaning the malicious software executes before a developer even begins using the package in their project. The worm then searches the developer’s system for tokens and access keys from npm, GitHub, or cloud services like AWS, Azure, and Google Cloud, sending these credentials to the attackers.
The most dangerous aspect is the malware’s self-propagation. Once the worm has a valid npm token, it logs in as that developer, finds all their packages, injects malicious code into each one, and publishes new trojanized versions. Analyses show tens of thousands of repositories where stolen credentials and attack artifacts have appeared, along with hundreds of compromised packages including those associated with Zapier, ENS, Postman, and AsyncAPI.
This attack illustrates the real risks associated with supply chains. Zapier, as one of the leading automation platforms, depends on an extensive array of open-source packages that connect developers and users worldwide.
Evil Twin Attack in Practice: Australian Man Sentenced to Over Seven Years for Rogue Wi-Fi Network
A man from Australia was sentenced to seven years and four months in prison for operating an “evil twin” Wi-Fi network during flights and at various airports across Australia. The attacker used a portable Wi-Fi Pineapple device to create fake access points with names identical to legitimate airport networks. People who connected to these fake networks were redirected to phishing pages where their social media login credentials were stolen.
The man used these stolen credentials to access women’s accounts, monitor their communications, and steal private images and videos. The Australian Federal Police (AFP) found thousands of intimate materials and other stolen data during their forensic investigation.
The man pleaded guilty to several offenses, including unauthorized data access, theft, interception of electronic communications, and tampering with evidence. The AFP warned the public about the risks of using public Wi-Fi networks and recommends using VPNs, strong passwords, and disabling automatic Wi-Fi connections.
GreyNoise IP Check: Find Out if Your Network Is Working for Hackers
GreyNoise Labs has introduced a free tool called GreyNoise IP Check, which allows users to verify whether their IP address has been involved in malicious scanning activities, such as those performed by botnets or proxy networks. This tool is designed to minimize network disruption while detecting whether devices on the network have been compromised or exploited for malicious operations.
Over the past year, there has been an increase in the use of proxy networks that often turn home internet connections into exit points for other people’s traffic. These networks are created through malware installed by users or by compromising unpatched devices on home networks.
The scanner returns three types of results: Clean (no malicious activity detected), Malicious/Suspicious (the IP address has shown suspicious behavior and may require further investigation), and Common Business Service (IP addresses belonging to VPNs, corporate networks, or cloud service providers).
The platform also offers a 90-day timeline of suspicious activity related to the IP address, helping users identify when an infection may have occurred. For technically savvy users, GreyNoise provides an unauthenticated JSON API for integration into scripts or monitoring systems.