Kernel, a company that runs browser sessions on hosted servers, says an attacker used a leaked API key to get into its systems, and it believes the attacker was working through ARTEX, an AI-driven penetration-testing platform. Security vendors have tied the same tool to espionage campaigns against more than 200 organizations.
What happened
In a notice dated November 4, 2026, Kernel (kernel.sh) said it detected a security incident on November 2. Its timeline says the attacker first used a leaked API key on October 24, then on October 27 used it to get into a staff login at Kernel’s logging provider and download data. Kernel says that data included a small number of Kernel API keys and third-party secrets for deployed apps.
On November 3, Kernel says the same party used a compromised credential for its hosting provider account to reach four GPU servers and start a cryptominer. Three of the servers ran Kernel browser sessions; the fourth was a development machine that Kernel says held no customer data. Kernel says it revoked credentials, blocked IP addresses, took the servers offline and rotated keys, and that 0.28% of its customers were affected. It describes the attacker as financially motivated and says it believes ARTEX served as their command center. That attribution is Kernel’s belief, not an independently confirmed finding.
Why ARTEX matters
According to Cybersecurity Dive (November 2), CrowdStrike reported that a group it tracks as Vapor Panda hit more than 200 organizations in 14 countries starting October 17, including foreign ministries, a Middle East interior ministry, and defense, pharmaceutical and telecom targets. Microsoft describes ARTEX as an open-source pen-testing framework that works with different AI models. SOCRadar (November 5), summarizing vendor research, says it uses coordinated AI agents for reconnaissance, exploitation, privilege escalation and data theft, and that it works with several model backends.
Google and Mandiant also report that several threat actors are using open-source pentest frameworks, ARTEX among them, to automate lateral movement and database exploitation. There is no single ARTEX vulnerability to patch. SOCRadar’s own summary is “no patch, only hardening.”
Who is affected
Kernel customers should watch for Kernel’s direct notice; the company says it contacted organizations whose data was exposed. More broadly, any team that keeps long-lived API keys, cloud credentials or third-party secrets in logging, hosting or deployment tools is the kind of target this activity favors.
What to do
The first two items follow Mandiant’s published advice as relayed by Cybersecurity Dive; the rest is our analysis.
- Reduce internet exposure of administrative and application-management interfaces.
- Restrict identities and keys to the minimum access they need, and use scoped API keys where a provider offers them.
- Rotate long-lived keys and secrets, especially any that appear in logs, exports or deployment tooling.
- Review who can sign in to logging, monitoring and hosting accounts, and require strong multi-factor authentication.
- Alert on unexpected GPU or compute usage, which was how this attacker’s cryptominer would show up.
What we still do not know
Kernel has not said how the original API key leaked or how the logging-provider and hosting credentials were compromised. It has not given an absolute customer count, and it does not say whether data in the browser sessions on the three servers was accessed. The sources also disagree on whether Vapor Panda overlaps with Microsoft’s Storm-0940, so we do not treat them as the same group. No evidence has been published linking the Kernel attacker to the espionage campaigns beyond the shared tool.
Sources: Kernel public notice of security incident (Nov. 4, 2026); Cybersecurity Dive (Nov. 2, 2026); SOCRadar (Nov. 5, 2026).
Leave a Reply