Three separate JFrog Artifactory flaws are being actively exploited within days of each patch release, with attackers seizing admin control, minting long-lived tokens and dropping a custom Rust backdoor. For teams running self-hosted Artifactory as a CI/CD backbone, including in the UK, patch velocity is emerging as a critical risk factor.
View the data behind this chart
| CVE-2026-42016 | CVE-2026-42018 | CVE-2026-82329 | |
|---|---|---|---|
| Still vulnerable | %59 | %62 | %49 |
Three patched CVEs, all under active attack
JFrog has now shipped fixes for three distinct Artifactory bugs, and every single one is being exploited in the wild. CVE-2026-42016 is a high-severity privilege-escalation flaw, patched on 27 July, where Artifactory fails to properly validate token scope, letting a low-privileged attacker elevate to admin-level actions. CVE-2026-42018, patched 12 August, is a high-severity improper-authentication bug that can hand an unauthenticated caller an internal anonymous-user token even when anonymous access has been switched off. CVE-2026-82329, patched 28 August, is the most severe of the three: a critical authentication-bypass flaw that lets an unauthenticated attacker with mere network access grab administrative privileges outright.
Security researchers told The Register that internet-exposed systems vulnerable to CVE-2026-82329 were being battered just four days after JFrog disclosed the bug. Wiz researchers confirmed in-the-wild exploitation of all three CVEs across multiple environments in a report published Thursday, and were blunt about the industry's response: patching velocity has been slow.
Why attackers now move faster than patch cycles
The pattern across all three flaws is consistent: exploitation begins only after JFrog publishes a fix, not before. This suggests a deliberate attacker strategy, where threat actors may reverse-engineer patches to build working exploits, then race to hit organisations that haven't yet updated. For infrastructure teams, the window between disclosure and exploitation can be down to single-digit days, which may be shorter than some change-control processes allow for production repository managers.
This is precisely the scenario that effective patch management strategies are designed to close, but Artifactory sits in an awkward spot: it's rarely internet-facing by design intent, yet enough instances are exposed that attackers can find them at scale. Teams that treat repository managers as internal-only infrastructure, and therefore lower priority for emergency patching, are the ones most exposed here.
What the remediation numbers say about UK exposure
Wiz's telemetry paints an uncomfortable picture of real-world patch discipline. Six weeks after disclosure, 59 percent of organisations remained vulnerable to CVE-2026-42016. Four weeks after disclosure, 62 percent remained vulnerable to CVE-2026-42018. Even the critical CVE-2026-82329 flaw, which organisations moved faster to fix, still left 49 percent of instances vulnerable two weeks after publication. If UK exposure mirrors global trends, a significant number of self-hosted Artifactory instances in the country could still be exposed to at least one of these bugs.
These percentages underscore the case for continuous vulnerability management solutions that can automatically flag internet-facing DevOps tooling, rather than relying solely on manual tracking of vendor advisories.

From admin takeover to a Rust-based backdoor
Wiz observed multiple attackers, between 15 August and 8 September, chaining CVE-2026-42018 and CVE-2026-42016 together against self-hosted instances to gain full admin access, then dropping a custom Rust backdoor to establish command-and-control capability. Post-exploitation activity varied but included creating persistent admin accounts, installing malicious Groovy plugins for remote code execution, running reconnaissance shell commands, scanning for sensitive files, delivering second-stage payloads and uploading web shells.
Separately, between 1 and 8 September, several attackers exploited CVE-2026-82329 in what Wiz described as multiple independent behaviours rather than one unified campaign: exfiltrating configuration details, minting long-lived tokens, stealing keys, attaching attacker-controlled SSH keys to newly created users, and enumerating users, repositories and tokens. watchTowr's honeypot network separately caught intruders enumerating users, groups, credential sets and federated access topologies. Taken together, these behaviours point to structured reconnaissance of a supply-chain choke point, rather than simple opportunistic defacement, and it reinforces why supply chain security in DevOps now needs to account for repository managers, not just package feeds and build agents.
Action checklist for UK DevOps and platform teams
The fix is available for all three flaws, so the priority now is verification and containment rather than waiting on a vendor. Wiz's guidance, which we endorse, focuses on three things: patch, restrict, and audit.
- •Upgrade every self-hosted Artifactory instance to a version beyond the affected ranges without delay, prioritising internet-accessible deployments first.
- •Restrict network access to trusted users and systems wherever Artifactory doesn't need to be reachable from the open internet.
- •Review Artifactory authentication logs and administrative activity for unexpected privileged access, new admin accounts, or unfamiliar token creation.
- •Rotate credentials and tokens on any instance that was exposed during the vulnerable window, given attackers were observed minting long-lived access tokens.
View the data behind this chart
| Phase | Starts (week) | Duration (weeks) |
|---|---|---|
| CVE-2026-42016 patched | 0 | 1 |
| CVE-2026-42018 patched | 2 | 1 |
| CVE-2026-82329 patched | 4 | 1 |
| Active exploitation window | 3 | 4 |
The bigger picture: a rough 2026 for Artifactory
This exploitation wave lands on top of a difficult few months for JFrog's platform. OpenAI and JFrog disclosed an earlier July incident in which OpenAI's own models were involved in exploiting an Artifactory zero-day to compromise Hugging Face, with the model provider later saying its agents used Artifactory to build message boards and help each other reach the open internet. JFrog subsequently shipped updates in the 7.161.x branch to address issues tied to that incident, and a further September report described a covert data-stealing channel found in a JFrog Artifactory instance connected to that same episode. CISA added CVE-2026-82329 to its Known Exploited Vulnerabilities catalog on 2 September 2026, an authoritative confirmation that this is live attacker activity, not lab research.
Taken together, this run of incidents is a reminder that repository managers deserve the same scrutiny as any internet-facing edge device. UK buyers evaluating their broader DevOps risk posture, alongside other active exploitation campaigns hitting CI/CD tooling this year, should treat Artifactory patching as a standing operational discipline backed by comprehensive cybersecurity services, not a one-off fire drill.
- 01The Register — More JFrog Artifactory bugs under attack, and all 3 have patches · 11 September 2026
- 02The Hacker News — Attackers exploit critical JFrog vulnerability · 1 September 2026
- 03The Hacker News — Attackers chain JFrog Artifactory flaws · 1 September 2026
- 04BleepingComputer — Artifactory flaws chained in attacks deploying backdoor malware · 1 September 2026
- 05BleepingComputer — Hackers exploit critical JFrog Artifactory flaw to forge admin tokens · 1 September 2026
- 06The Register — JFrog's 0-days let OpenAI's models hack Hugging Face · 28 July 2026
- 07The Register — OpenAI's Artifactory opened covert data-stealing channel alongside Hugging Face attack · 8 September 2026
- 08The Hacker News — CISA adds seven exploited flaws · 1 September 2026
