A supply-chain attack against Aqua Security’s Trivy ecosystem let attackers publish malicious artifacts and hijack GitHub Action tags, turning a widely used security scanner into a credential stealer. Reporting indicates the intrusion began with abuse of a misconfigured GitHub Actions workflow and theft of privileged credentials, followed by incomplete containment that left residual access in place. Attackers then poisoned aquasecurity/trivy-action by force-updating 75 of 76 tags, compromised setup-trivy, and published a backdoored Trivy v0.69.4 release; later activity also pushed malicious Docker Hub images 0.69.5 and 0.69.6. The malware harvested GitHub tokens, cloud credentials, SSH keys, Kubernetes secrets, Docker configs, and other CI/CD data from runners and developer environments, encrypted the loot, and exfiltrated it to attacker-controlled infrastructure or fallback GitHub repositories such as tpcp-docs.
Researchers and vendor advisories linked the campaign to TeamPCP and described it as an expanding, multi-stage operation that also included a brief OpenVSX compromise of the Trivy VS Code extension, defacement of 44 repositories in Aqua Security’s internal aquasec-com GitHub organization, and follow-on compromises affecting Checkmarx tooling and the LiteLLM PyPI package. Aqua removed malicious artifacts, revoked tokens, restored safe references, and said commercial products were not affected, while GitHub and public advisories identified safe versions including Trivy 0.69.2/0.69.3, trivy-action 0.35.0, and setup-trivy 0.2.6. U.S. CISA added CVE-2026-33634 to the KEV catalog, and incident responders warned organizations that ran affected versions to assume full pipeline compromise, rotate all accessible secrets, audit workflow logs and GitHub activity, and pin GitHub Actions to immutable commit SHAs.

Mallory correlates global threat intelligence with your attack surface — know if you’re exposed before adversaries strike.
27 events from the most recent confirmed update back to the earliest known activity.
By 2026-04-06, reporting described a separate compromise of the widely used Axios npm package in which malicious releases delivered a cross-platform RAT through a fake dependency. The incident was attributed by Google and Microsoft to a North Korea-linked threat actor.
By 2026-04-03, CERT-EU confirmed that the European Commission’s Europa AWS hosting environment was breached through the Trivy supply-chain compromise tracked as CVE-2026-33634. The intrusion reportedly began with stolen AWS API keys on March 19, was detected on March 24, and led to exfiltration of 340 GB of data affecting 71 clients.
By 2026-04-01, reporting said Cisco's build and development environment had been breached through a malicious GitHub Action associated with the Trivy campaign, leading to theft of credentials and cloning of hundreds of repositories. Cisco said it was aware of the issue and later stated it had not found evidence of impact to customers, products, or services.
On 2026-03-31, Mercor confirmed it was affected by the LiteLLM compromise and said it had contained and remediated the incident while continuing forensic investigation. Subsequent reports tied the case to extortion claims and alleged theft of large volumes of company data.
By 2026-03-31, ownCloud disclosed that its build infrastructure had been affected through CVE-2026-33634 linked to the Trivy supply-chain compromise. The company said customer data and source code were not impacted.
By 2026-03-30, Databricks said it was investigating an alleged compromise potentially linked to TeamPCP-harvested credentials from the broader supply-chain campaign. Later reporting indicated Databricks had not found evidence of internal impact and requested more information.
On 2026-03-27, TeamPCP reportedly used valid Telnyx publishing credentials to upload malicious releases to PyPI. The packages executed code on import and concealed a second-stage payload in a WAV file, extending the campaign into automated build and production environments.
On 2026-03-27, Kudelski Security published forensic analysis of the Trivy compromise describing separate GitHub Action and binary variants, including a persistent backdoor installed via a systemd user service in non-CI environments. The report said affected clients experienced theft of AWS and CI/CD credentials followed by attacker reconnaissance across 29 AWS regions, and concluded that data in 24 scanned S3 buckets should be treated as potentially compromised.
On 2026-03-27, CISA added CVE-2026-33634 to its Known Exploited Vulnerabilities catalog and set an April 9, 2026 remediation deadline for U.S. federal civilian agencies. The KEV entry confirmed active exploitation and reinforced guidance to remove affected artifacts and rotate secrets.
On 2026-03-26, the Trivy ecosystem compromise was formalized as CVE-2026-33634, describing embedded malicious code distributed through compromised Trivy artifacts and GitHub Actions. Public advisories emphasized that exploitation could expose tokens, SSH keys, cloud credentials, and other CI/CD secrets.
By 2026-03-24, Mandiant said the Trivy supply-chain attack had already impacted more than 1,000 SaaS environments and warned of aggressive downstream extortion and follow-on compromises. Aqua acknowledged that incomplete containment after the earlier incident likely allowed the attacker to retain or regain access.
Later on 2026-03-24, the malicious LiteLLM versions were removed from PyPI, BerriAI froze releases, and external incident response support was engaged. Users were told to treat any environment running the affected versions as fully compromised and rotate credentials.
On 2026-03-24, attackers used stolen publishing credentials to release backdoored LiteLLM versions 1.82.7 and 1.82.8 on PyPI. Version 1.82.8 included a malicious .pth file that executed automatically on Python interpreter startup, enabling credential theft, persistence, and Kubernetes lateral movement.
Follow-up analysis published on 2026-03-26 found the Checkmarx ast-github-action compromise was broader than first reported, with all 91 published tags overwritten on March 23. Defenders were advised to review any use during the exposure window and move to safe version v2.3.33.
On 2026-03-23, TeamPCP expanded the campaign to Checkmarx, poisoning KICS-related GitHub Action tags and publishing malicious ast-results and cx-dev-assist plugin versions to OpenVSX. Checkmarx later said the VS Code Marketplace was unaffected and released clean replacement versions after revoking affected tags and securing access.
On 2026-03-22, attackers used a stolen long-lived service account token, likely Argon-DevOps-Mgt, to rename and deface all 44 repositories in Aqua Security's internal aquasec-com GitHub organization within minutes. The incident raised concern that proprietary code, CI/CD configurations, and internal documentation were exposed.
On 2026-03-22, additional malicious Trivy Docker Hub image tags 0.69.5 and 0.69.6 were published without matching GitHub releases or tags. Analysis linked them to the same TeamPCP infostealer infrastructure, expanding exposure beyond GitHub Actions and the v0.69.4 release.
On 2026-03-21, and reflected in later reporting, Aqua issued GitHub advisory GHSA-69fq-xp46-6x23 covering the March Trivy compromise. The notice documented affected binaries, Docker images, and GitHub Actions, identified safe versions, and attributed the renewed access to incomplete credential rotation after the late-February attack.
On 2026-03-20, Aqua removed malicious artifacts where possible, restored safe references, and advised users to move to known-safe versions such as Trivy 0.69.2/0.69.3, trivy-action 0.35.0, and setup-trivy 0.2.6. The company also recommended pinning GitHub Actions by full commit SHA and auditing workflows for compromise.
The malicious tag rewrites were detected around 19:15 UTC on 2026-03-19 and publicly confirmed on 2026-03-20. Aqua and researchers warned that any workflows using affected tags should be treated as fully compromised and that secrets needed immediate rotation.
On 2026-03-19, StepSecurity published trivy-compromise-scanner, a CLI tool that audits GitHub Actions workflow logs for evidence of the aquasecurity/trivy supply-chain compromise during the March 19–20 exposure window. The tool supports repository- and organization-wide scans, checks compromised action references and SHAs, and outputs findings in JSON, CSV, and summary table formats.
On 2026-03-19, attackers with retained write access compromised Aqua Security's Trivy ecosystem again, publishing a malicious Trivy v0.69.4 release and force-updating most aquasecurity/trivy-action tags plus setup-trivy tags to malicious commits. The payload acted as an infostealer targeting CI/CD runners and developer environments.
By 2026-03-03, StepSecurity and others reported that hackerbot-claw had opened more than a dozen pull requests across multiple repositories, achieved code execution in several targets, and stole tokens from projects including awesome-go and Aqua Security's Trivy. The reporting tied the Trivy OpenVSX incident to this broader CI/CD exploitation wave.
By 2026-02-28/2026-03-01, Aqua published and later revised GitHub Security Advisory GHSA-8mr6-gf9x-j8qg and stated it was under active attack affecting Trivy-related assets. Reporting later indicated the initial credential changes on March 1 did not fully evict the attacker.
On 2026-02-28, a former Aqua employee associated with the OpenVSX publisher account revoked the publishing token and removed the affected Trivy extension versions from OpenVSX. This limited the exposure window to roughly one day.
On February 27–28, 2026, suspicious OpenVSX versions 1.8.12 and 1.8.13 of Aqua's Trivy VS Code extension were published under the aquasecurityofficial.trivy-vulnerability-scanner namespace. The injected code silently launched local AI coding assistant CLIs with non-interactive flags and attempted broad system inspection and possible exfiltration.
Activity attributed to the autonomous GitHub account hackerbot-claw began around 2026-02-20/21, targeting misconfigured GitHub Actions workflows in open-source repositories. The campaign used unsafe patterns such as pull_request_target with untrusted code to gain code execution and steal credentials.
Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.
50 references tracked. Mallory keeps watching after this page renders.
github.com
Open sourcecrowdstrike.com
Open sourceiansresearch.com
Open sourcevulnu.com
Open sourcesecurityaffairs.com
Open sourcecybersecuritynews.com
Open sourcecybersecuritynews.com
Open sourcemitiga.io
Open sourceMap indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.