Threat reportThreat IntelligenceTL-2026-1109
Nebula — AI-Integrated Open-Source Penetration Testing Tool (BerylliumSec) — Dual-Use Tool Tracking, No CVE/Active Exploitation
Nebula is an open-source, LLM-integrated penetration testing CLI tool from BerylliumSec that wraps OpenAI, Llama-3.1-8B-Instruct, Mistral-7B-Instruct, and DeepSeek-R1-Distill-Llama-8B models around Nmap, Metasploit, CrackMapExec, and OWASP ZAP to automate recon, exploit suggestion, and engagement documentation. This record tracks Nebula as a dual-use offensive-automation tool for defender awareness — it is not a vulnerability, exploit, or confirmed malicious campaign, and carries no CVE/CVSS.
Key facts
| Identifiers | TL-2026-1109 |
|---|---|
| Also known as | Nebula AI, nebula-ai, BerylliumSec Nebula |
| Severity | INFO |
| Affected products | BerylliumSec Nebula (nebula-ai); BerylliumSec Nebula Pro; BerylliumSec Deep Application Profiler (DAP) |
| Tools | CrackMapExec - S0488, Deep Application Profiler (DAP), DeepSeek-R1-Distill-Llama-8B, Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.2, Nebula, Nebula Pro, Nmap, OWASP ZAP, Ollama, metasploit |
| MITRE ATT&CK techniques | T1003 OS Credential Dumping, T1018 Remote System Discovery, T1021 Remote Services, T1046 Network Service Discovery, T1049 System Network Connections Discovery, T1059.006 Python, T1068 Exploitation for Privilege Escalation, T1087 Account Discovery, T1110 Brute Force, T1113 Screen Capture, T1190 Exploit Public-Facing Application, T1570 Lateral Tool Transfer and 6 more in the attack flow |
| Motivation | UNKNOWN |
| Detection rules | 9 (1 shown free) |
| Sources | 15 cited sources |
Detection coverage
As of 2026-07-05, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-1109 across Splunk SPL, Microsoft KQL and Sigma, covering 23 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.
Free detection: Unsanctioned nebula-ai / neutron-ai / nebula-watcher Package Install
Detects installation or upgrade of the BerylliumSec Nebula AI-pentest tooling family ('nebula-ai', 'neutron-ai', 'nebula-watcher') via pip on endpoints not designated as approved red-team/pentest jump hosts. Presence of this package is a strong forensic indicator that AI-assisted offensive automation (Nmap/Metasploit/CrackMapExec/OWASP ZAP orchestration) has been provisioned on the host, which may represent authorized purple-team use or an unsanctioned/unauthorized deployment.
- Languages: Splunk SPL, Microsoft KQL, Sigma
- ATT&CK: T1588.002, T1587.001
- Data sources: sysmon_process_creation, linux_auditd, bash_history
- Severity: medium
- Confidence: high
Splunk SPL
index=endpoint (sourcetype=XmlWinEventLog:Microsoft-Windows-Sysmon/Operational EventCode=1 OR sourcetype=linux:audit OR sourcetype=bash_history)
| eval cmd=lower(coalesce(CommandLine, ExecCommand, command))
| where like(cmd, "%pip install nebula-ai%") OR like(cmd, "%pip install neutron-ai%") OR like(cmd, "%pip install nebula-watcher%") OR like(cmd, "%pip install --upgrade nebula-ai%") OR like(cmd, "%pip3 install nebula-ai%")
| stats count min(_time) as first_seen max(_time) as last_seen values(cmd) as commands by host, user
| lookup approved_pentest_hosts.csv host OUTPUT is_approved
| where isnull(is_approved) OR is_approved="false"
| convert ctime(first_seen) ctime(last_seen)
| sort -last_seenMicrosoft KQL
DeviceProcessEvents
| where Timestamp > ago(24h)
| where ProcessCommandLine has_any ("pip install nebula-ai", "pip install neutron-ai", "pip install nebula-watcher", "pip3 install nebula-ai")
| extend IsApprovedHost = DeviceName in ("jump-host-approved-01", "jump-host-approved-02")
| where IsApprovedHost == false
| project Timestamp, DeviceName, AccountName, ProcessCommandLine, InitiatingProcessFileName
| order by Timestamp descSigma
title: Nebula-style AI-Pentest-Tool Package Install (nebula-ai family)
id: 8f2b1c4e-0a11-4e2a-9c3d-1a2b3c4d5e6f
status: experimental
description: Detects pip installation of BerylliumSec's Nebula AI-integrated pentest CLI family (nebula-ai, neutron-ai, nebula-watcher), a dual-use LLM-driven offensive automation toolset that wraps Nmap/Metasploit/CrackMapExec/OWASP ZAP.
references:
- https://github.com/berylliumsec/nebula
- https://pypi.org/project/nebula-ai/
- https://cybersecuritynews.com/nebula-ai-penetration-testing/
author: Threadlinqs Team
date: 2026-07-05
tags:
- attack.resource-development
- attack.t1588.002
- attack.t1587.001
logsource:
category: process_creation
product: linux
detection:
selection:
CommandLine|contains:
- 'pip install nebula-ai'
- 'pip install neutron-ai'
- 'pip install nebula-watcher'
- 'pip3 install nebula-ai'
condition: selection
falsepositives:
- Authorized purple-team or internal red-team engagement provisioning an approved AI-pentest jump host
- Security research/lab environments intentionally testing Nebula
level: mediumKnown false positives:
- Approved red-team/purple-team engagement installing Nebula on a sanctioned jump host
- Security research lab intentionally evaluating the tool
Tuning notes: Maintain an approved-host allowlist (lookup table or DeviceName list) and suppress alerts from it; expand the CommandLine match list if BerylliumSec renames the PyPI package or ships additional companion tools.
- OPENAI_API_KEY Environment Set on Host Also Invoking Offensive Security Tooling · unlock with TL Creds
- Account Discovery Immediately Followed by Multi-Host Credential Brute Force · unlock with TL Creds
- Nebula Default Log Directory and Screenshot/Engagement-Note Artifact Creation · unlock with TL Creds
- Ollama Local LLM Model Pull Consistent with Nebula AI-Pentest Provisioning · unlock with TL Creds
- CrackMapExec Credential-Dumping Module Invocation (LSASS/SAM/NTDS) · unlock with TL Creds
- DNS Query to OpenAI API Endpoint Initiated by Offensive Security Tooling Process Tree · unlock with TL Creds
- Rapid AI-Chained Discovery-to-Brute-Force Sequence (CrackMapExec-style, Multi-Host) · unlock with TL Creds
- Lateral Tool Transfer via SMB/SSH Immediately Following AI-Assisted Recon/Brute-Force Chain · unlock with TL Creds
Indicators of compromise
As of 2026-07-05, Threadlinqs Intelligence tracks 23 indicators of compromise for TL-2026-1109.
Technical analysis
Nebula is an open-source, AI-integrated penetration testing CLI tool developed by BerylliumSec (github.com/berylliumsec/nebula), distributed via PyPI as the 'nebula-ai' package. Dated PyPI release history shows the 1.0.9 beta line already active by 2024-01-18 (version 1.0.9b37) and continuing through 2024-03-20 (version 1.0.9b39), with the tool later rewritten onto the 2.0 beta line by 2025-02-04 (version 2.0.0b4). It wraps large language models — OpenAI's API-accessible models, Meta's Llama-3.1-8B-Instruct, Mistral AI's Mistral-7B-Instruct-v0.2, and DeepSeek-R1-Distill-Llama-8B — around a pentester's terminal workflow, running local inference through Ollama (CPU/GPU capable) or cloud inference via an OPENAI_API_KEY environment variable. Minimum system requirements are 16GB RAM and Python 3.10-3.13.9, with Docker deployment available using X11 forwarding for GUI features.
Nebula does not replace existing offensive tooling; it 'works alongside any CLI-invokable security utility,' with documented integrations for Nmap, Metasploit, CrackMapExec, and OWASP ZAP. Users toggle between AI and Terminal modes (or prefix commands with '!') to get AI-powered internet-search-augmented context, real-time exploitation suggestions derived from parsing terminal tool output, automated note-taking and finding categorization (independently observed by Ostorlab to map findings to CWE and NIST standards), built-in screenshot capture/annotation, and a status feed panel refreshing every five minutes. Nebula Pro, a commercial tier built on the open-source Nebula 2.0 codebase, adds an 'Autonomous Mode' that charts and chains multi-tool attack paths with human oversight, plus a code-analysis capability and multi-analyst collaboration with audit trails. BerylliumSec separately publishes the Deep Application Profiler (DAP), a neural-network- and vector-database-backed malware analysis web service/API (listed on Microsoft AppSource) aimed at zero-day detection in unmanaged-code executables up to roughly 0.5 MB, and a related terminal assistant called 'neutron' (github.com/berylliumsec/neutron).
This is a TOOL-RELEASE TRACKING record, not a vulnerability, exploit, or active campaign: there is no associated CVE, CVSS score, confirmed malicious deployment, or threat-actor attribution. It is documented here because Nebula is explicitly dual-use — the same LLM-driven reconnaissance, vulnerability-scanning, exploit-suggestion, credential-attack (via CrackMapExec), and lateral-movement (via CrackMapExec/Metasploit) automation that benefits authorized pentesters could, in principle, be adopted by an intrusion actor to accelerate equivalent stages of an unauthorized attack. Independent reviews (Darknet.org.uk, Ostorlab, Starlog, Spark42.tech) consistently describe Nebula as an AI-assisted terminal assistant that keeps a human in the loop by default (autonomous multi-step chaining is opt-in and Pro-only) and caution that LLM-generated exploit suggestions can be confidently wrong — citing non-existent CVEs, syntactically broken payloads, or OS-mismatched privilege-escalation techniques — so any AI-suggested exploit should be treated as an unverified hypothesis requiring validation.
Defensive relevance: SOC/blue teams should be able to recognize Nebula's footprint — the 'nebula-ai' PyPI package, its default '~/.local/share/nebula/logs' log path, Ollama model-pull commands (e.g., 'ollama pull mistral'), and OPENAI_API_KEY usage — to distinguish authorized red-team/pentest engagement activity from potential unauthorized use of the same tool, and to correlate Nebula-driven Nmap/CrackMapExec/Metasploit/ZAP activity with existing detections built for those underlying utilities.
MITRE ATT&CK techniques by tactic:
Credential Access
T1003 OS Credential Dumping; T1110 Brute Force
Discovery
T1018 Remote System Discovery; T1046 Network Service Discovery; T1049 System Network Connections Discovery; T1087 Account Discovery
Lateral Movement
T1021 Remote Services; T1570 Lateral Tool Transfer
Execution
Privilege Escalation
T1068 Exploitation for Privilege Escalation
Collection
Initial Access
T1190 Exploit Public-Facing Application
Resource Development
T1587.001 Malware; T1588.002 Tool; T1588.006 Vulnerabilities
Reconnaissance
T1592 Gather Victim Host Information; T1595.002 Vulnerability Scanning; T1596 Search Open Technical Databases
Remediation
Immediate actions
- Inventory hosts for the 'nebula-ai' PyPI package, Ollama model pulls, and the default '~/.local/share/nebula/logs' log directory to identify unsanctioned Nebula installations
- Restrict and monitor outbound calls authenticated via OPENAI_API_KEY from non-approved pentest jump hosts
Longer-term hardening
- Maintain an approved-tool allowlist for internal red-team/purple-team engagements and require sign-off before AI-assisted offensive tooling (Nebula, Nebula Pro, or similar) is used against production systems
- Extend existing detections for Nmap, Metasploit, CrackMapExec, and OWASP ZAP to correlate with AI-assistant wrapper activity, since Nebula automates invocation of these tools rather than replacing them
- Treat AI-generated exploit or privilege-escalation suggestions from any LLM-assisted tool as an unverified hypothesis requiring manual validation before execution, given documented hallucination risk
Affected and fixed versions
- BerylliumSec — Nebula (nebula-ai)
Vulnerable versions: 2.0.0b4 (2025-02-04 PyPI release); 2.0 open-source line - BerylliumSec — Nebula Pro
Vulnerable versions: Preview/commercial tier adding Autonomous Mode and Code Analysis - BerylliumSec — Deep Application Profiler (DAP)
Vulnerable versions: Web service / API, Microsoft AppSource listing
Timeline
- Earliest confirmed dated PyPI release of the 'nebula-ai' package, version 1.0.9b37, showing the 1.0.9 beta line of Nebula was already under active development in early 2024.
- Nebula version 1.0.9b39 is published to PyPI, continuing the 1.0.9 beta release line prior to the later 2.0 rewrite.
- Nebula version 2.0.0b4 is published to PyPI as the 'nebula-ai' package, reflecting the jump to the 2.0 beta line and active continued beta-stage development.
- Darknet.org.uk publishes 'Nebula – Autonomous AI Pentesting Tool,' describing Nebula's integrations with Nmap, CrackMapExec, and OWASP ZAP and noting it is an AI-assisted terminal tool rather than a fully autonomous pentester (exact publish day approximate, sourced to April 2025).
- Spark42.tech publishes 'Top 10 Open-Source AI Agent Penetration Testing Projects,' listing Nebula among leading open-source AI pentesting agent projects.
- Ostorlab publishes '8 Open-Source AI Pentest Tools for Security Teams in 2026,' benchmarking Nebula and peer tools against a banking web application and noting Nebula's AI-assisted findings map to CWE and NIST standards.
- Starlog publishes 'Nebula: The AI-Powered Pentesting Assistant That Learns Your Workflow,' profiling Nebula's workflow-adaptive AI assistant features.
- Cyber Security News publishes 'New Nebula AI-Integrated Penetration Testing Tool Empowers Attackers to Uncover Weaknesses' by Guru Baran, the article that triggered this tool-tracking record; it confirms LLM backends and capabilities with no associated CVE or exploitation activity.
- BerylliumSec publishes its own comparison blog post, 'AI-Powered Penetration Testing: Nebula in Focus and How It Stacks Up Against the Rest,' positioning Nebula against competing open-source AI pentesting agents.
Sources cited
- Nebula – AI-Powered Penetration Testing Assistant (PyPI project page) pypi.org · Vendor report · 2025-02-04
- AI-Powered Penetration Testing: Nebula in Focus and How It Stacks Up Against the Rest berylliumsec.com · Vendor report · 2026-07-05
- Deep Application Profiler (DAP) — Microsoft AppSource listing Microsoft · Vendor report · 2026-07-05
- nebula-ai 1.0.9b37 (PyPI release history) pypi.org · Vendor report · 2024-01-18
- nebula-ai 1.0.9b39 (PyPI release history) pypi.org · Vendor report · 2024-03-20
- nebula-ai — PyPI Package Security Analysis (Socket.dev) socket.dev · Research · 2026-07-05
- neutron-ai — PyPI Package Security Analysis (Socket.dev) socket.dev · Research · 2026-07-05
- berylliumsec/nebula (GitHub repository) GitHub · Code repository · 2026-07-05
- berylliumsec/nebula_watcher (GitHub repository) GitHub · Code repository · 2026-07-05
- berylliumsec/neutron (GitHub repository) GitHub · Code repository · 2026-07-05
- Nebula – Autonomous AI Pentesting Tool darknet.org.uk · Blog · 2025-04-01
- Top 10 Open-Source AI Agent Penetration Testing Projects blog.spark42.tech · Blog · 2025-11-27
- 8 Open-Source AI Pentest Tools for Security Teams in 2026 blog.ostorlab.co · Blog · 2026-01-30
- Nebula: The AI-Powered Pentesting Assistant That Learns Your Workflow starlog.is · Blog · 2026-05-08
- New Nebula AI-Integrated Penetration Testing Tool Empowers Attackers to Uncover Weaknesses cybersecuritynews.com · Article · 2026-07-03
Previous and next in threat intel
Drafted by the Threadlinqs research pipeline from 15 sources; published ; last updated —.