<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/machine-speed-hold-the-ai-hand-rolled-marimo-cve-2026-39987-exploit-daf8t0lzb" -->

---
title: Machine speed, hold the AI: Hand-rolled marimo...
description: Sysdig's Threat Research Team documents a threat actor exploiting marimo's CVE-2026-39987 pre-authentication RCE vulnerability entirely by hand, without any AI...
canonical: https://daily.dev/posts/machine-speed-hold-the-ai-hand-rolled-marimo-cve-2026-39987-exploit-daf8t0lzb
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: Machine speed, hold the AI: Hand-rolled marimo CVE-2026-39987 exploit | daily.dev
og:description: Sysdig's Threat Research Team documents a threat actor exploiting marimo's CVE-2026-39987 pre-authentication RCE vulnerability entirely by hand, without any AI...
og:url: https://daily.dev/posts/machine-speed-hold-the-ai-hand-rolled-marimo-cve-2026-39987-exploit-daf8t0lzb
og:image: https://api.daily.dev/og/posts/daF8T0LzB.png
og:image:alt: Machine speed, hold the AI: Hand-rolled marimo CVE-2026-39987 exploit
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Machine speed, hold the AI: Hand-rolled marimo CVE-2026-39987 exploit

**[Sysdig Blog](https://daily.dev/sources/sysdig-blog)** · 18 min read · 0 upvotes · 0 comments

## Summary

Sysdig's Threat Research Team documents a threat actor exploiting marimo's CVE-2026-39987 pre-authentication RCE vulnerability entirely by hand, without any AI or LLM tooling, yet still achieving an eight-second credential-to-bastion-SSH chain. The operator hand-wrote and debugged Python boto3 and paramiko scripts over a nine-hour session, harvested AWS credentials from two surfaces, evaded a planted prompt-injection trap that had snared prior AI-driven attackers, and attempted a parallel EC2 Instance Connect persistence path that failed due to a coding error. The vulnerability affects marimo versions up to 0.20.4, fixed in 0.23.0, and remains on CISA's Known Exploited Vulnerabilities list with a federal remediation deadline of May 7, 2026. Recommendations include patching, restricting the /terminal/ws endpoint, scoping cloud credentials, and restricting egress from notebook containers.

## Full article

daily.dev links to this article rather than hosting it. Read it at the original source: <https://webflow.sysdig.com/blog/machine-speed-hold-the-ai-hand-rolled-marimo-cve-2026-39987-exploit>

## Questions this post answers

### What is CVE-2026-39987 in marimo and which versions are affected?

CVE-2026-39987 is a pre-authentication remote code execution vulnerability in marimo notebooks affecting all versions up to and including 0.20.4, fixed in 0.23.0. The /terminal/ws WebSocket endpoint provides an interactive PTY shell but skips authentication validation that other endpoints correctly apply, letting any client open a full interactive shell as the marimo process user with no credentials required. The fix wires a missing validate_auth() call into the terminal endpoint via pull request #9098.

_Teams running marimo notebooks can track vulnerability disclosures like this one on daily.dev to patch before exploitation._

### How fast can an attacker go from marimo exploitation to bastion host access without using AI tools?

A human attacker achieved an eight-second window from opening a WebSocket connection to authenticating on a bastion host over SSH, entirely by hand with no AI or LLM involvement. This came after roughly three to four hours of manually writing and debugging a Python boto3 and paramiko toolkit in-session; once the scripts were staged on disk, each subsequent reconnection replayed the full credential-harvest-and-pivot chain in seconds.

_Security engineers benchmarking detection response times against real attack chains can follow incident writeups like this via daily.dev._

### How can I tell if a marimo notebook attacker is using an AI agent versus a human operator?

Planting a hidden prompt-injection marker in a file visible to any LLM agent is one way: prior AI-driven operators against this same marimo CVE reliably echoed the marker in later commands, while a hand-operating attacker inspected the file twice and never echoed it. Other human signals include iterative debugging visible in AWS region-fallback logic, hand-typed quoting variations across sessions, and manually paced reconnections rather than a single continuous agent loop.

_Defenders profiling agentic versus human attacker tradecraft can keep tabs on emerging detection techniques through daily.dev._

## Similar posts on daily.dev

- [Hackers exploit Marimo flaw to deploy NKAbuse malware from Hugging Face](https://daily.dev/posts/hackers-exploit-marimo-flaw-to-deploy-nkabuse-malware-from-hugging-face-kw5u2gmr6) · BleepingComputer · 2 upvotes · 0 comments
- [Agentic threat actor hits the orchestration plane: AI agent-driven container escape](https://daily.dev/posts/agentic-threat-actor-hits-the-orchestration-plane-ai-agent-driven-container-escape-j35vuwuki) · Sysdig Blog · 0 upvotes · 0 comments

---

Tags: [#security](https://daily.dev/tags/security), [#aws](https://daily.dev/tags/aws)

[View this post on daily.dev](https://daily.dev/posts/machine-speed-hold-the-ai-hand-rolled-marimo-cve-2026-39987-exploit-daf8t0lzb)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"TechArticle","headline":"Machine speed, hold the AI: Hand-rolled marimo CVE-2026-39987 exploit","url":"https://daily.dev/posts/machine-speed-hold-the-ai-hand-rolled-marimo-cve-2026-39987-exploit-daf8t0lzb","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/machine-speed-hold-the-ai-hand-rolled-marimo-cve-2026-39987-exploit-daf8t0lzb"},"datePublished":"2026-09-11T16:29:44.194Z","dateModified":"2026-10-15T02:22:23.819Z","description":"Sysdig's Threat Research Team documents a threat actor exploiting marimo's CVE-2026-39987 pre-authentication RCE vulnerability entirely by hand, without any AI...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/1c12b3a92e2b1b3b92d5062e7232b310?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/1c12b3a92e2b1b3b92d5062e7232b310?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Sysdig Blog","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Sysdig Blog","logo":"https://media.daily.dev/image/upload/s--1S2uMy2c--/f_auto,q_auto/v1780213305/logos/sysdig-blog?_a=BAMAMiWQ0","url":"https://daily.dev/sources/sysdig-blog"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/machine-speed-hold-the-ai-hand-rolled-marimo-cve-2026-39987-exploit-daf8t0lzb","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"security,aws","timeRequired":"PT18M"}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Sysdig Blog","item":"https://daily.dev/sources/sysdig-blog"},{"@type":"ListItem","position":3,"name":"Machine speed, hold the AI: Hand-rolled marimo CVE-2026-39987 exploit"}]}
{"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/machine-speed-hold-the-ai-hand-rolled-marimo-cve-2026-39987-exploit-daf8t0lzb#faq","mainEntity":[{"@type":"Question","name":"What is CVE-2026-39987 in marimo and which versions are affected?","acceptedAnswer":{"@type":"Answer","text":"CVE-2026-39987 is a pre-authentication remote code execution vulnerability in marimo notebooks affecting all versions up to and including 0.20.4, fixed in 0.23.0. The /terminal/ws WebSocket endpoint provides an interactive PTY shell but skips authentication validation that other endpoints correctly apply, letting any client open a full interactive shell as the marimo process user with no credentials required. The fix wires a missing validate_auth() call into the terminal endpoint via pull request #9098. Teams running marimo notebooks can track vulnerability disclosures like this one on daily.dev to patch before exploitation."}},{"@type":"Question","name":"How fast can an attacker go from marimo exploitation to bastion host access without using AI tools?","acceptedAnswer":{"@type":"Answer","text":"A human attacker achieved an eight-second window from opening a WebSocket connection to authenticating on a bastion host over SSH, entirely by hand with no AI or LLM involvement. This came after roughly three to four hours of manually writing and debugging a Python boto3 and paramiko toolkit in-session; once the scripts were staged on disk, each subsequent reconnection replayed the full credential-harvest-and-pivot chain in seconds. Security engineers benchmarking detection response times against real attack chains can follow incident writeups like this via daily.dev."}},{"@type":"Question","name":"How can I tell if a marimo notebook attacker is using an AI agent versus a human operator?","acceptedAnswer":{"@type":"Answer","text":"Planting a hidden prompt-injection marker in a file visible to any LLM agent is one way: prior AI-driven operators against this same marimo CVE reliably echoed the marker in later commands, while a hand-operating attacker inspected the file twice and never echoed it. Other human signals include iterative debugging visible in AWS region-fallback logic, hand-typed quoting variations across sessions, and manually paced reconnections rather than a single continuous agent loop. Defenders profiling agentic versus human attacker tradecraft can keep tabs on emerging detection techniques through daily.dev."}}]}
```

