JetBrains has told users of Cadence, its cloud service for running machine-learning workloads from PyCharm, to rotate every credential and secret they have ever used on the platform after disclosing that attackers exploited an unpatched flaw in its own TeamCity software to break into Cadence's infrastructure.
In a security notice posted to its blog, JetBrains said threat actors exploited CVE-2026-63077, a critical deserialization bug in TeamCity carrying a near-maximum CVSS severity score of 9.8, to run commands on the Cadence server with the privileges of the TeamCity process itself. The company said "activity identified from August 8, 2026" continued undetected until August 23, when JetBrains found the intrusion and took the server offline the next day.
What attackers reached was extensive: usernames, real names, email addresses, login timestamps and IP addresses tied to Cadence accounts; a full 2024 backup of the server containing stored credentials and configuration files; AWS IAM credentials, including some belonging to JetBrains employees; contents of Amazon S3 buckets inside JetBrains' own AWS accounts; and source code that had synced over from users' PyCharm projects. Because Cadence executes code on users' behalf, JetBrains is now telling customers to treat every input and output from past Cadence runs as potentially compromised.
An unusually blunt apology
JetBrains did not try to soften what went wrong. "The server should have been patched as part of our response to the vulnerability, but it was not," the company wrote. "We sincerely apologize for this failure and the impact it may have on you." CVE-2026-63077 had been public since July and was added to the U.S. Cybersecurity and Infrastructure Security Agency's Known Exploited Vulnerabilities catalog on August 5 — three days before the intrusion began — meaning the Cadence server sat unpatched against a flaw federal regulators had already flagged as being actively exploited in the wild, as The Hacker News reported in its account of the breach.
JetBrains has invalidated all Cadence plugin access tokens and is urging affected users to revoke and rotate AWS keys, API tokens and any other secrets referenced in Cadence executions, then audit their own repositories for unauthorized changes. The company said its investigation is continuing and that it will publish further findings as they emerge.
The episode is a reminder that AI-development tooling has become as attractive a target as the models themselves: Cadence exists specifically to let PyCharm users offload GPU-heavy training and inference jobs to the cloud, which means a single compromised orchestration server can expose the credentials, code and cloud infrastructure of a large number of unrelated projects at once.
