basicsecurity.net
Proof, not just disclosure.
Threats / Marimo / CVE-2026-39987
CVE-2026-39987 · EUVD no mirror located · GCVE no mirror located Verified 2026-06-22

Marimo vulnerability

Marimo contains a pre-authorization remote code execution vulnerability allowing unauthenticated attackers to gain shell access and execute arbitrary system commands.

Verdict

Today item — known-exploited.

An unauthenticated remote attacker can exploit missing authorization controls to execute arbitrary code on affected Marimo instances, leading to full system compromise without requiring valid credentials.

CISA KEV Yes · 2026-04-233EPSS 0.95645 (verify live)4Exploit Weaponized · public PoC5
01

Is it exploitable?

— the evidence, ranked above the score
Exploit available
Fully weaponized — public exploit code is cataloged for this vulnerability.We link the existence of the exploit; we do not host or redistribute payloads.
Reported exploitation
7 independent public reports of in-the-wild exploitation are cataloged.Distinct reporting sources (vendor, incident response, government); open them for the underlying claims.
Exploited in the wild
Listed in the CISA Known Exploited Vulnerabilities catalog (added 2026-04-23).
CISA KEV ↗Confirmed
Probability (EPSS)
EPSS 0.95645 — modeled likelihood of exploitation activity.EPSS is a daily-changing model output — open the source for today's value.
Severity / affected
Affected: Marimo, Marimo. Confirm exact fixed builds in the vendor advisory.
NVD ↗Reported
Weakness (CWE)
Mapped to CWE-306 Missing Authentication — weakness family: Authentication.CWE assignment from the public NVD record; the weakness class drives how the flaw is exploited.
NVD ↗Reported
WeaknessCWE-306 · Missing AuthenticationAuthentication
02

Who’s exploiting it?

— attribution turns risk into urgency
Attribution not established

No confirmed (advisory-backed) threat-actor attribution is established for this record. Absence of a named actor is not absence of compromise — see Coverage & confidence.

03

Why it matters

— the attack path, told twice: adversary, then board
1

Front door — unauthenticated access narrative 1

Attacker
I identify a Marimo instance exposed on the network and send a crafted request to trigger code execution without authentication.
Business
The organization's Marimo deployment becomes an entry point for unauthorized access to critical systems.
2

Keys to the kingdom — privilege/identity takeover narrative 2

Attacker
I execute arbitrary system commands through the vulnerability to establish persistent shell access on the compromised host.
Business
Attackers gain foothold for lateral movement, data exfiltration, and further infrastructure compromise.
3

Lateral reach — past segmentation narrative 3

Attacker
I leverage the shell access to escalate privileges, access sensitive data, or deploy additional malware.
Business
Confidentiality, integrity, and availability of business systems and data are severely compromised.
04

What to do

— defensible action
  • Remediate per the vendor advisory — confirm the fixed build for your version and verify exposure.1
Say it to the boardA vulnerability with this evidence profile is a defensible budget line, not a backlog ticket — fund the change against the proof above.
05

Coverage & confidence

— what we know, and what we don’t

Established (cited)

  • KEV listing (CISA)
  • EPSS probability (FIRST)
  • Weaponized exploit available (VulnCheck)
  • 7 reported-exploitation source(s)
  • CWE weakness mapping (NVD)
  • Public exploit availability
  • Catalogued by GitHub_M (CNA)
  • Coverage gaps — stated, not hidden

  • No EUVD / GCVE mirror in feed — single-authority dependency for the identifier.
  • EPSS & exposure are time-varying; verify live at the source.
  • Threat-actor attribution not established from feed data — absence of a name is not absence of compromise.
  • No finder/reporter credit recorded in the public CVE entry — the work behind this find is unattributed.
  • Disclosure & credit2
    Catalogued by GitHub_MCNA
    Credited with finding itNo finder named in the public CVE record — the work behind this find is unattributed.