What happened
Replaying the Checkmarx KICS March 2026 intrusion
- 1
Access to Checkmarx's publishing accounts
March 19, 2026Credentials that Checkmarx believes came from the Trivy supply chain attack give access to its GitHub repositories; how the publishing accounts were obtained is not stated.
- 2
Poisoned actions and extensions
March 23, 2026; the action tags from 12:58 to 16:50 UTCAll 35 KICS tags and the AST tags are repointed to backdated attacker commits, two Open VSX versions are published, and workflows that name a poisoned tag run the setup script first.
- 3
Secrets taken on the runners
March 23, 2026; stated from the payload's codeThe script reads runner memory, environment variables and SSH keys, asks the EC2 metadata service for role credentials and, with a mounted token, plants a privileged pod and a backdoor on each node.
- 4
The harvest sent out
March 23, 2026The encrypted harvest is posted to a look-alike domain, with a release in the victim's GitHub account as fallback; Checkmarx reports no evidence of successful exfiltration from customer environments.
- 5
Checkmarx's own estate and the second wave
March 30 to May 10, 2026Checkmarx's repository data is taken on March 30, its AWS credentials see initial reconnaissance, and kept access publishes a second wave from April 22.
From 12:58 to 16:50 UTC on March 23, 2026, all 35 KICS action tags and the AST action tags served a credential stealer to the workflows that named them, and two Open VSX extension versions carried a stealer for longer. As Wiz reads its code, the script reads runner secrets, asks the EC2 metadata service for role credentials and, with a mounted token and enough permissions, creates privileged pods that install a host backdoor; no source names a victim, runner or cluster where it did so, and Checkmarx reports no evidence of successful exfiltration from customer environments. Data from Checkmarx's GitHub repositories was taken on March 30 and published on April 25, a second wave followed on April 22 without overriding known safe versions, and a limited number of Checkmarx workstations were infected; Mandiant found initial reconnaissance of Checkmarx AWS credentials, while the AWS production environment, Checkmarx One SaaS and CxSAST on-premises were not affected.
Where cloud-provider controls fit
The tags, the publishing accounts and the extensions live in GitHub and in registries outside every cloud, so GitHub's own settings act first: an Actions policy that requires full commit SHAs refuses the poisoned tags in your workflows, while tag rulesets, immutable releases, a token policy and an IP allow list bound the publishing side. Where the script runs on your runners, IAM roles federated through OIDC in place of stored keys, a metadata hop limit of 1, role credentials bound to your VPC, a DNS allow list and pod admission rules bound what a job hands it. No source names a victim whose secrets were taken; Checkmarx's repository data was taken on March 30, while its AWS production environment and Checkmarx One SaaS were not impacted, per Mandiant's review.
Could this have been prevented?
In hindsight, not as a whole. 1 control would have refused a step outright, and 21 controls would have limited one. 11 steps have no cloud-provider control. The counts below are control assessments, not attack stages: several can address the same step. The same replay with the controls in place before the incident shows where each phase would have been refused or bounded.
- 1
- Prevented
- 21
- Contained
- 11
- Not covered
How it could have been prevented
The same replay with the controls in place. Green marks where a step is refused, blue where it is bounded, and red where no control reaches it. Open a row to read the control, why it acts, and its caveat.
Replaying the Checkmarx KICS March 2026 intrusion
1 · Access to Checkmarx's publishing accounts
2 Contained1 Not covered2 · Poisoned actions and extensions
1 Prevented2 Contained4 Not covered3 · Secrets taken on the runners
8 Contained3 Not covered9 on other clouds4 · The harvest sent out
3 Contained1 Not covered3 on other clouds5 · Checkmarx's own estate and the second wave
6 Contained2 Not covered3 on other cloudsKey facts
- From 12:58 to 16:50 UTC on March 23, 2026, all 35 KICS action tags pointed to individually crafted, backdated attacker commits published through the cx-plugins-releases service account; the AST action tags were overwritten too, all 91 of them per an independent researcher cited by SANS ISC on March 26
- On March 23, 2026, ast-results 2.53.0 and cx-dev-assist 1.7.0 were published to Open VSX through one account, 12 seconds apart; ReversingLabs counted about 36,000 downloads of ast-results and about 500 installs of cx-dev-assist as of March 27, without saying whether those cover only the malicious versions
- Per Wiz's analysis of the code, the script read Runner.Worker memory, environment variables and SSH keys, queried the EC2 instance metadata service over IMDSv1 and IMDSv2, created a privileged pod in kube-system on each node when a mounted token allowed it, and posted the encrypted harvest to a look-alike domain, with a release in the victim's GitHub account as fallback
- Data was exfiltrated from Checkmarx's GitHub repositories on March 30, 2026 with credentials compromised in the March wave, and a cybercriminal group published it on the dark web on April 25
- On April 22, 2026, kept access published malicious KICS Docker Hub tags, ast-github-action 2.3.35 and four extension versions on the Microsoft marketplace and Open VSX; stolen npm tokens published @bitwarden/cli 2026.4.0 on April 23, with about 334 downloads in a 93-minute window per Trend Micro, and a modified Jenkins AST plugin was available from May 9 to 10
- Mandiant's review, as Checkmarx reported it, found initial reconnaissance of Checkmarx AWS credentials and a limited number of infected workstations; the AWS production environment was not impacted, Checkmarx One SaaS was not accessed, and Checkmarx declared the incident fully contained on July 6, 2026
Sources
Every fact in this breakdown comes from the public reporting below. The replay is a model of that reporting, not a reconstruction of the environment it describes.
- 1.Update: Ongoing Checkmarx Supply Chain Security IncidentCheckmarx · Jul 6, 2026 · Vendor reportAffected artifacts, exposure windows, safe versions, access vector as believed, repository data theft, second wave, Jenkins plugin, Mandiant findings, customer impact, remediation
- 2.KICS GitHub Action Compromised: TeamPCP Strikes Again in Supply Chain AttackWiz · Mar 24, 2026 · ResearchTag hijack window, publishing identity, payload steps, Kubernetes persistence, exfiltration, Open VSX payload, takedown and reinstatement
- 3.TeamPCP expands: Supply chain compromise spreads from Trivy to Checkmarx GitHub ActionsSysdig · Mar 23, 2026 · ResearchAST action compromise, observed IMDS access, observed encrypted upload, runtime detection
- 4.TeamPCP Supply Chain Campaign: Update 001 - Checkmarx Scope Wider Than Reported, CISA KEV Entry, and Detection Tools AvailableSANS Internet Storm Center · Mar 26, 2026 · ResearchAST tag count, AST remediation window, action.yml change, backdated commits
- 5.Checkmarx KICS GitHub Action Compromised: Malware Injected in All Git TagsStepSecurity · Mar 23, 2026 · ResearchTag state, discovery, payload targets as reported, guidance
- 6.The TeamPCP supply chain attack evolvesReversingLabs · Mar 27, 2026 · ResearchOpen VSX versions, download counts, availability, payload in general
- 7.Analysing TeamPCP's Supply Chain Attacks: Checkmarx KICS and elementary-data in CI/CD Credential TheftTrend Micro · May 13, 2026 · ResearchSecond-wave payload, Bitwarden CLI follow-on, access vector as assessed
- 8.Checkmarx Confirms Data Stolen in Supply Chain AttackSecurityWeek · Apr 29, 2026 · NewsRepository data theft, extortion claims, response
The estate in the replay is illustrative and drawn at the level of attack stages; the victim's real environment is not public. Verdicts describe what a control does when it is in place before the incident. They are not an assessment of the victim's security program.