CVE
HKUDS Nanobot AI sandbox escape via caller-selected shell
Only affects nanobot 0.2.1 or later before 0.4.1 with sandboxed exec enabled, restrictToWorkspace=true, and sandbox=bwrap. Fixed in 0.4.1.
Written by a model from the sources below. Verify against them.
CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:C/C:H/I:H/A:H
- Exploited
- No known exploitation
- Fix
- Fixed in 0.4.1
- Who is exposed
- Specific setups only
Do first: Check whether your setup matches what the bug needs. Sandboxed exec is enabled, reproduced with restrictToWorkspace=true and sandbox=bwrap. nanobot version 0.2.1 or later but before 0.4.1. The attacker can induce the agent to create an executable in the workspace and select it as the shell.
- Last modified
- Whether it needs action now or is a reference record, by fixed rules, with why. Last checked is when cve.now last re-read it.What do these mean? →
- Reference
- Last checked
- FIRST's estimate of the chance attackers exploit it in the next 30 days, updated daily.What do these mean? →
- 0.37% · 30th percentile
- A 0-100 score for how much a CVE matters across the industry: exploitation, exposure, how widely the software is used, severity and whether a fix exists.What do these mean? →
- 28/100 · provisional
- How many real-world setups can hit the bug, from the vendor's advisory and CISA data. It does not know your setup.What do these mean? →
- Specific setups only
- The share of the registry's downloads of each affected package in the window that were versions inside the affected ranges. Downloads, not users or installs: CI runs and mirrors inflate the counts.
- 80% of nanobot-ai downloads last week were affected versions (14.7K).PyPI, per-version counts from ClickPy, 18.5K downloads in the last 7 days, read Nov 14, 2026
- Whether working attack code is public (Exploit-DB, Metasploit, Nuclei or GitHub). Public code makes attacks easier.What do these mean? →
- None on Exploit-DB, Metasploit or GitHub
- Exploitation: none
- Automatable: no
- Technical impact: total
Summary
HKUDS's nanobot (package nanobot-ai) lets a command run outside its sandbox when the caller selects the shell executable, so an attacker can run a program with the same operating-system permissions as the nanobot process. An attacker with low-privilege access over the network, and no user action, can do this by getting the agent to create an executable in the workspace and selecting it as the shell.
Original description
In nanobot, when sandboxed exec is enabled, the shell executable used to run commands can be selected by the caller and is launched before the sandbox is applied. An attacker who can induce the agent to create an executable in the workspace and select it as the shell can cause that executable to run outside the sandbox (e.g. Linux bubblewrap), with the operating-system permissions of the nanobot process.
This issue affects nanobot (nanobot-ai) from 0.2.1 before 0.4.1.
Who should check
Teams running nanobot-ai 0.2.1 to before 0.4.1 with sandboxed exec (bwrap) enabled and an authenticated WebSocket user able to prompt the agent.
“Nanobot 0.2.1 < 0.4.1”
cve.now Index
- Exploitation0/40No signal
- Exposure8/25Specific setups only, reachable over the network
- Reach10/20Affects PyPI packages
- Severity9.5/10CVSS 8.5, total technical impact per CISA
- No patch0/5No signal
What would change the score
- Would score about 78 if added to CISA KEV
- Would score about 73 if CISA's SSVC rates exploitation active
- Would score about 35 if CISA's SSVC rates it automatable
Impact
- Widely used
nanobot-ai has 17.1k weekly downloads.Source: pypistats.org
5 packages depend on nanobot-ai.Source: deps.dev
What to do
Check whether your setup matches what the bug needs.
Sandboxed exec is enabled, reproduced with restrictToWorkspace=true and sandbox=bwrap. nanobot version 0.2.1 or later but before 0.4.1. The attacker can induce the agent to create an executable in the workspace and select it as the shell.Upgrade to a fixed version: 0.4.1.
Check your systems with an agent
Paste this into a coding or ops agent. It checks your systems read-only and reports a table per asset.
Check whether my systems are affected by CVE-2026-121864 (Nanobot Sandbox Escape via Caller-Selected Shell Executable), using the cve.now data below.
Rules: work read-only. Do not run exploit code, proof-of-concept scripts or scanners that send attack payloads, and do not change any configuration, package, image or deployment. Report findings only.
What cve.now has for CVE-2026-121864 (entry last updated 2026-11-17):
Affected:
- PyPI package nanobot-ai: from 0.2.1, fixed in 0.4.1
Fixed versions: 0.4.1.
Conditions (all must be true for the bug to be hit):
- Sandboxed exec is enabled, reproduced with restrictToWorkspace=true and sandbox=bwrap. (source: https://www.tenable.com/security/research/tra-2026-68)
- nanobot version 0.2.1 or later but before 0.4.1. (source: https://www.tenable.com/security/research/tra-2026-68)
- The attacker can induce the agent to create an executable in the workspace and select it as the shell. (source: https://github.com/HKUDS/nanobot/security/advisories/GHSA-5v87-m222-qr94)
Setup: unknown whether the bug needs a non-default feature or setting. Check the advisory.
Attacker needs: network access, low-privilege login, no user action.
CISA KEV: CVE-2026-121864 is not listed as of cve.now's last read.
This KEV status is for CVE-2026-121864 only. If you also check another CVE, look up its own KEV status; never assume it from this one.
How to check:
- Packages (PyPI): find every lockfile in each project and run the stack.check operation of the cve.now MCP server (execute) on each one. Compare the resolved versions, not the ranges a manifest declares, with the affected versions above. Include transitive dependencies.
Verify against the linked advisories before concluding, and cite the date you read each one:
- https://www.tenable.com/security/research/tra-2026-68
- https://github.com/HKUDS/nanobot/security/advisories/GHSA-5v87-m222-qr94
- https://github.com/HKUDS/nanobot/releases/tag/v0.4.1
Do not conclude an asset is unaffected unless you checked it and the evidence covers both its version and the conditions above. If you could not check something, mark it unknown and say what is missing.
Report a table with one row per asset and these columns: Asset | Version or build | Status (affected, fixed, not applicable, unknown) | Compensating controls | Exposure (internet-facing, internal, none) | Evidence | Next action.
Entry: https://cve.now/CVE-2026-121864
For follow-up questions, use the cve.now MCP server at https://mcp.cve.now/mcp (no authentication), for example execute with name cve.get and arguments {"id": "CVE-2026-121864"}.Timeline
- CVE published
- First seen by cve.now
- CISA SSVC assessment
Exploitation: none · Automatable: no · Technical impact: total
- Record last updated
Affected versions
- nanobot-aiPyPI
From 0.2.1 · fixed in 0.4.1
Fixed in 0.4.1 (released Nov 6, 2026, 11 days before disclosure).
The latest release, 0.4.1, is not affected.
Evidence 5
Vendor advisory · 1
Establishes: Affected and fixed versions.
- Advisory
CVE record · 2
Establishes: The identifier and description.
- Record
- Record
Other links · 2
- Repository
- Link
Corrections
None so far.
Something wrong on this page? Report an error. Every fix is listed here.
Questions about CVE-2026-121864
Is CVE-2026-121864 being exploited?
It is not in CISA's Known Exploited Vulnerabilities catalog, and CISA's SSVC assessment does not mark exploitation as active. That does not mean it has not been attacked: vendors and researchers can report attacks before CISA lists them. No public exploit code is tracked.Source: cve.now feed data
What does this mean?
The shell that nanobot runs can be an executable the attacker induces the agent to create in the workspace and then selects as the shell. That executable runs outside the bwrap sandbox, with the nanobot process's operating-system permissions, so it can reach any host resource available to that account.Source: tenable.com, github.com
Which versions are affected by CVE-2026-121864?
nanobot-ai (PyPI): from 0.2.1 before 0.4.1.Source: cve.now feed data
Am I affected, and how do I check?
Versions 0.2.1 or later but before 0.4.1 are affected when sandboxed exec is enabled with restrictToWorkspace=true and sandbox=bwrap. To check, look at the installed nanobot-ai version and at the restrictToWorkspace and sandbox settings in the configuration.Source: tenable.com, github.com
How do I fix CVE-2026-121864?
Upgrade nanobot-ai to 0.4.1 or later using the existing installation method, then restart the gateway.Source: tenable.com, github.com