In short
Mixpanel stated an attacker accessed a part of its methods and exported customer-identifiable metadata.
OpenAI stated no prompts, API keys, cost info, or authentication tokens had been concerned.
Each corporations reviewed the incident, notified affected customers, and outlined new safety steps.
A breach at analytics supplier Mixpanel earlier this month uncovered account names, e mail addresses, and browser places for some customers of OpenAI’s API, the AI big confirmed Wednesday, elevating issues that cybercriminals might use the stolen metadata in focused phishing makes an attempt.
In response to Mixpanel, on November 8, an unknown attacker gained entry to a part of its methods and exported a dataset containing customer-identifiable metadata and analytics info. The stolen information included usernames, e mail addresses, approximate browser-based location, working system, and browser particulars.
OpenAI stated the breach didn’t embody customers’ prompts, API keys, cost info, or authentication tokens.
Solely information from customers who accessed OpenAI’s tech through the API—aka, through exterior apps powered by GPT—was leaked, the corporate stated. In different phrases, in case you entry the ChatGPT chatbot immediately from OpenAI’s web site, you then will not be impacted right here.
“As a part of our safety investigation, we eliminated Mixpanel from our manufacturing providers, reviewed the affected datasets, and are working intently with Mixpanel and different companions to totally perceive the incident and its scope,” OpenAI stated in a press release.
Based in 2009, the San Francisco-based Mixpanel is a product analytics platform used to trace consumer conduct throughout internet and cellular purposes. The corporate stated it detected the “smishing” marketing campaign, and after an preliminary investigation and response, alerted OpenAI the subsequent day.
“We’re dedicated to transparency, and are notifying all impacted clients and customers,” OpenAI stated. “We additionally maintain our companions and distributors accountable for the very best bar for safety and privateness of their providers.”
Smishing is a kind of phishing assault performed by means of SMS messages. In response to an October report by infrastructure administration firm Spacelift, smishing accounted for 39% of all cellular threats in 2024.
Mixpanel stated it secured affected accounts, revoked lively classes, rotated compromised credentials, and blocked malicious IP addresses. The corporate additionally reset worker passwords, employed exterior cybersecurity companies, and reviewed authentication, session, and export logs.
After the breach, Mixpanel stated it started notifying impacted clients concerning the incident.
“When you have not heard from us immediately, you weren’t impacted,” Mixpanel CEO Jen Taylor stated in a press release. “We proceed to prioritize safety as a core tenet of our firm, merchandise, and providers. We’re dedicated to supporting our clients and speaking transparently about this incident.”
Regardless of Mixpanel’s reporting of the incident to OpenAI, the ChatGPT developer stated it was chopping ties with the analytics agency. “After reviewing this incident, OpenAI has terminated its use of Mixpanel,” they wrote.
Some OpenAI clients took to social media to precise frustration with the revelation {that a} third-party service had entry to their info.
“I am not very joyful about this. […] Why did they need to move on my title and e mail deal with to Mixpanel?” one consumer wrote on X. “I’m only a hobbyist attempting to make small experiments.”
“OpenAI sending names and emails to a 3rd occasion analytics platform (Mixpanel) feels wildly irresponsible,” one other wrote.
OpenAI and Mixpanel didn’t instantly reply to requests for remark by Decrypt.
Typically Clever E-newsletter
A weekly AI journey narrated by Gen, a generative AI mannequin.
