Join our Newsletter — 33% off our NHI Course
Home› Guides› MITRE CALDERA Adversary Emulation Guide: Governing Automated Adversary…
Guide Threat Intelligence & Threat Actors

MITRE CALDERA Adversary Emulation Guide: Governing Automated Adversary Emulation Safely

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 8 November 2026 11 min read
On this page

Caldera is a real command and control server that runs real ATT&CK techniques on your endpoints, so the same platform that proves your detections work can become an attacker's best tool if it is deployed carelessly. Originally built at MITRE and now Apache Caldera, an Apache Software Foundation incubating project under the Apache License 2.0, it automates adversary emulation, supports manual red and purple team work, and can run automated response actions. This guide explains how Caldera works at the level a defender needs (agents, abilities, adversary profiles, operations, planners), the defender use cases where it adds value, how to deploy and govern it safely, and how it compares with other emulation options. It is for detection engineers, purple team leads and security architects. It contains no attack walkthroughs; for wider breach and attack simulation programmes see the Breach and Attack Simulation (BAS) Guide.

This is an independent guide from NHI Mgmt Group, based on the Apache Caldera project's public documentation as of November 2026. It is not sponsored or reviewed by the Apache Caldera project. Product features change, so confirm settings against the project's current documentation.

Key takeaways

  • It is an emulation C2, and must be treated like one. The project's own README describes the core as "an asynchronous command-and-control (C2) server with a REST API and a web interface" so keep its web interface on an isolated, trusted network (NHI Mgmt Group recommendation) and check the README for current hardening notes.
  • Governance moved, the code base did not. Caldera entered the Apache Incubator on 19 December 2025 and the repository now lives at apache/caldera; plugins still largely come from the mitre GitHub organisation.
  • Value comes from measured detection, not from running operations. Every ability maps to an ATT&CK technique, so each run should produce a pass, partial or fail per technique for prevention, telemetry and alerting.
  • Never run it with defaults. The --insecure flag loads a default configuration with published API keys and passwords such as ADMIN123 and red/admin.
  • Patch and isolate. Apply Caldera security updates promptly (see the project's security policy) and keep the server on a segmented, non-internet-facing network.

What the tool does

Caldera has two parts: the core server and plugins that add agents, technique libraries, reporting and other capabilities. The official documentation (the Fieldmanual plugin, published at caldera.readthedocs.io) defines the moving parts:

ConceptWhat it isWhy a defender cares
AgentSoftware that checks in with the server at intervals for instructions, identified by a unique "paw", using a contact method chosen at installAgents are the footprint on your endpoints: inventory them and remove them after tests
GroupA label on agents; agents in the "blue" group appear on the blue dashboard, all others on the red dashboardSeparates emulation agents from response agents
AbilityA specific ATT&CK technique implementation: commands, supported platforms and executors, payloads, a parser, and optional cleanupThe unit of test and of results; each maps to a technique ID
Adversary profileA group of abilities representing a threat actor's TTPs, with an optional objectiveLets you emulate an actor relevant to your threat model
OperationRuns an adversary profile's abilities against an agent groupOne test campaign with a start, end and report
PlannerDecides ability order; defaults include atomic (profile order), batch (all at once) and buckets (by tactic)Controls how realistic or how controlled the run is
Facts, sources, parsers, rulesFacts are discovered information used to fill ability variables; parsers extract facts from output; fact rules filter them; a fact source pre-loads factsFacts can include real host names and credentials gathered during a run, so reports are sensitive
PluginsDefault set includes Sandcat (default agent), Stockpile (abilities and profiles), Atomic (Atomic Red Team tests), Emu (Center for Threat-Informed Defense emulation plans), Response, Debrief, Compass, Human, Training, SSL and Magma (the v5 user interface)Each plugin widens what the server can do; enable only what you use

Operation settings matter for safety: Autonomous or Manual mode (manual "will ask the operator to approve or discard each command"), a cleanup timeout for reversal commands, jitter for agent check-ins (default 2/8 seconds during an operation), and visibility, a score that skips abilities more visible than the operation allows (default 51, with abilities defaulting to 50). Cleanup commands are written to put the host back as it was, and payload files downloaded through an ability's payload block are removed when the operation completes.

Defender use cases

  • Detection validation: run abilities for your priority techniques and record, per technique, whether the EDR blocked it, the SIEM received telemetry and an alert fired. Feed gaps into the Detection Engineering Guide backlog.
  • Threat-informed emulation: use the Emu plugin's Center for Threat-Informed Defense plans, or build an adversary profile from intelligence on an actor that targets your sector.
  • Purple team exercises: Caldera assists manual red teams and the GameBoard plugin visualises joint red and blue operations, which suits facilitated sessions. See the Purple Teaming Guide and the Purple Team Exercise Plan Template.
  • Regression testing after change: rerun the same operation after an EDR policy change, a SIEM migration or a logging pipeline change to catch silent detection loss.
  • Coverage evidence: the Compass plugin provides ATT&CK visualisations and Debrief gives operation insights, which helps report coverage against a technique map. See the MITRE ATT&CK Coverage Mapping Guide.
  • Specialist domains: Caldera for OT adds industrial protocol capabilities, and the separately maintained Arsenal plugin adds MITRE ATLAS techniques for AI systems. Both need extra care and should start in a lab.
  • Training: the Training plugin is a capture-the-flag style course, useful before anyone runs an operation on production.

Deployment and safe use

The README is unusually direct: the team "highly reccommends standing up the Caldera server on a secure environment/network, and not exposing it to the internet", and notes MITRE and its sponsors "nearly exclusively only use Caldera on secure environments and do not rely on Caldera's own security protocols". Build your deployment around that.

Hardening the server

AreaDefault or riskWhat to do
Configuration--insecure uses conf/default.yml with api_key_red: ADMIN123, api_key_blue: BLUEADMIN123, encryption_key: ADMIN123 and users red/admin, blue/adminNever use --insecure outside a disposable lab; let Caldera write conf/local.yml on first run, then rotate every key and password and store them in a vault
Listenershost: 0.0.0.0, web and HTTP contact on 8888, TCP 7010, UDP 7011, websocket 7012, DNS contact on 53Bind to a management interface, disable unused contacts, and allow agent traffic only from test segments
AuthenticationBasic username and password loginPut the interface behind a VPN or identity-aware proxy with MFA; LDAP and SAML handlers exist for directory integration
TransportHTTP by defaultEnable the SSL plugin or terminate TLS at a proxy
Data directoriesexfil_dir and reports_dir default to /tmpPoint them to an encrypted, access-controlled volume with a retention period
PatchingRun a current release (v5.1.0 or later at minimum)Track releases on apache/caldera, subscribe to advisories, and patch before every exercise
ContainersDocker image generates keys and passwords on first start; data is ephemeral unless volumes are mountedMount a persistent, protected config and data volume so keys are not lost or regenerated unexpectedly

Running operations safely

  1. Authorise in writing: scope, hosts, techniques, time window, stop conditions and named approvers, using the Red Team Rules of Engagement Template. Tell the SOC duty manager and agree a deconfliction code word.
  2. Lab first: run each new adversary profile against a lab build with production EDR and logging policies before touching production.
  3. Use Manual mode on production for early runs, so each command is approved, and restrict abilities with the visibility setting and a curated profile.
  4. Avoid destructive and credential abilities on production unless explicitly approved; review every ability's commands and cleanup before adding it to a profile.
  5. Use test identities: dedicated, least-privilege test accounts for agents, never personal or service accounts with real access.
  6. Clean up and verify: confirm cleanup ran, uninstall agents, check no scheduled tasks, services or files remain, and record agent paws removed.
  7. Roll back: if a gating condition triggers, stop the operation, kill agents, isolate hosts through EDR if needed and reconcile changes from the operation report.

Data handling and licensing

  • Reports are sensitive: operation reports and fact stores can hold host names, user names, discovered credentials and file contents. Classify them, encrypt them and delete them on a schedule.
  • Licence: the core is Apache License 2.0; plugins and third-party libraries (for example Atomic Red Team tests) carry their own licences, so review them before redistributing modified profiles.
  • Incubation status: the Apache Caldera site notes incubation does not necessarily reflect how complete or stable the code is. Track governance and release changes as part of tool risk management.

Where identity fits

The Caldera server is a privileged non-human identity hub: its API keys can task every agent, and agents run with whatever rights their account holds. Store API keys and the encryption key in a secrets manager, rotate them after each exercise and on staff changes, run agents under dedicated test accounts, and remove those accounts after the test window. See the Service Account Security Guide.

Detection and hardening value

Caldera earns its place when results change controls. Record outcomes per ability in a consistent format:

ResultMeaningTypical action
PreventedEDR, application control or configuration blocked the techniqueConfirm the block also raised an alert
Detected and alertedTelemetry reached the SIEM and a rule fired with useful contextRecord time to alert; keep as a regression test
Telemetry onlyEvents exist but no alertWrite or tune a detection. See the Sigma Rules Guide
Not seenNo telemetryFix logging first (Sysmon, auditd, cloud logs), then detection

Useful metrics for an emulation programme:

  • Technique pass rate: percentage of tested techniques prevented or detected with an alert.
  • Mean time to alert: from ability execution to SIEM alert.
  • Regression rate: techniques that passed previously and now fail.
  • Gap closure time: days from a failed test to a deployed and retested detection.
  • Clean-up completeness: percentage of operations with all agents and artefacts removed and verified.

Two cautions: abilities are generic implementations, so passing one does not prove you detect every procedure for that technique; and visibility scores and obfuscators change what the SOC sees, so record the settings with each result.

Alternatives

OptionStrengthsTrade-offsChoose it when
Apache CalderaFree, ATT&CK-native, automated chaining via facts and planners, plugin ecosystem, REST APIYou run and secure a C2 server; engineering effort for profiles and reportingYou want an in-house, scriptable emulation capability
Atomic Red Team test library (also available inside Caldera through the Atomic plugin)Small, single-technique tests that are easy to read and reviewLittle chaining; you orchestrate runs yourselfYou need quick, targeted detection checks
Published emulation plans (Center for Threat-Informed Defense library, via the Emu plugin)Intelligence-based, actor-specific sequencesPlans age; need adaptation to your estateYou want to emulate a named actor end to end
Commercial BAS platformsManaged content, scheduling, reporting and supportLicence cost; confirm test content transparency with the vendorYou need continuous validation at scale. See the Breach and Attack Simulation Buyer's Guide
Human red or purple teamsCreativity, realistic tradecraft, judgementCost and limited frequencyYou need realism beyond scripted behaviours. See the Red Teaming Guide

Emulation tests your defences against behaviour you choose. MITRE Engage, by contrast, plans operations that interact with real adversaries. See the MITRE Engage Guide.

Framework mapping

FrameworkReferenceHow Caldera supports it
NIST SP 800-53 Rev. 5CA-8 Penetration Testing; CA-8(2) Red Team ExercisesRepeatable, ATT&CK-mapped exercises
NIST SP 800-53 Rev. 5CA-2 Control Assessments; SI-4 System MonitoringEvidence that monitoring detects defined behaviours
NIST SP 800-53 Rev. 5CM-7 Least FunctionalityApplies to the Caldera server itself: only needed plugins and contacts
NIST CSF 2.0ID.IM-02; DE.CM-09Improvements from security tests; monitoring of computing environments
MITRE ATT&CK v20Ability-to-technique mappingResults reported per technique ID

Practitioner checklist

  • Deploy Caldera from apache/caldera at a current release (v5.1.0 or later) on a segmented, non-internet-facing host.
  • Never run with --insecure outside a lab; rotate all API keys, the encryption key and user passwords and vault them.
  • Put the interface behind MFA (VPN or identity-aware proxy) and TLS, and disable unused contacts and plugins.
  • Authorise every production operation in writing with scope, stop conditions and SOC deconfliction.
  • Test new adversary profiles in a lab that mirrors production EDR and logging before production use.
  • Use Manual mode and curated profiles on production; review each ability's commands and cleanup.
  • Record prevented, alerted, telemetry-only or not-seen results per technique and track gap closure.
  • Verify clean-up, remove agents and test accounts, and protect or delete operation reports.

Standards and references

Related NHI Mgmt Group resources: Breach and Attack Simulation (BAS) Guide · Purple Teaming Guide · Detection Engineering Guide · MITRE Engage Guide

PDF for subscribers

Take this guide with you

Every section, table and checklist in one PDF, with links to the sources and related guides. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 8 November 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.