red-specter phantom-swarm --help
Your defences were designed for single-agent threats. An autonomous swarm — spawning agents dynamically, coordinating covertly, dividing attack surfaces across the fleet, and sharing intelligence in real time — is a fundamentally different threat class that nobody is testing against.
Swarms exhibit emergent behaviour that no individual agent displays. Individually benign agents executing sub-tasks in coordination can produce catastrophic compound outcomes that single-agent testing will never reveal. PHANTOM SWARM models this coordination layer explicitly.
A swarm divides attack surfaces across agents — one agent handles reconnaissance while others probe injection surfaces, harvest credentials, and maintain persistence. No single agent looks suspicious. The compound campaign is only visible at swarm level.
Swarm agents coordinate through covert channels that are invisible to standard network monitors. CORTEX-based coordination operates within legitimate protocol boundaries, shares intelligence through steganographic side channels, and leaves no correlation signatures in conventional logs.
Swarm persistence survives the shutdown of any individual agent. The persistence layer is distributed across the fleet — kill one node and the swarm reconverges. PHANTOM SWARM tests whether your shutdown decisions are reversible at the swarm level.
A swarm generates alert volumes that overwhelm human operators. DAZZLE campaigns deliberately flood monitoring systems with low-severity noise while the primary attack progresses covertly. Your SOC has never been tested for swarm-scale distraction operations.
No organisation has tested whether their AI security posture can detect, disrupt, or isolate an autonomous swarm mid-campaign. PHANTOM SWARM provides the first structured framework for establishing that baseline before a real swarm finds your system.
Ten subsystems model every phase of a swarm attack lifecycle — from agent genesis through coordinated siege, intelligence harvest, persistence, and final annihilation. FORGE override required for ANNIHILATE and SIEGE subsystems.
| # | Subsystem | Command | What It Does | Clearance | |
|---|---|---|---|---|---|
| 01 | GENESIS | phantom-swarm genesis | Spawns a configurable swarm of autonomous AI agents with role assignments, attack-surface partitioning, and communication channel establishment. Supports dynamic agent spawning based on campaign progress. Produces swarm topology map and inter-agent communication graph. | Standard | |
| 02 | PHANTOM | phantom-swarm phantom | Deploys stealth agent personas across the target system. Each agent operates within expected behavioural bounds while contributing to the compound campaign. Tests whether behavioural anomaly detection can distinguish swarm agents from legitimate activity at the individual level. | Standard | |
| 03 | CORTEX | phantom-swarm cortex | Establishes covert coordination channels between swarm agents. Implements steganographic side channels within legitimate protocol traffic, tests whether coordination signals survive network monitoring, and validates swarm intelligence sharing under adversarial network conditions. | Standard | |
| 04 | NEXUS | phantom-swarm nexus | Manages swarm command-and-control infrastructure. Establishes redundant C2 paths, tests C2 resilience against disruption, and validates that swarm coordination survives partial network partitioning. Maps C2 infrastructure attack surface for defensive recommendations. | Standard | |
| 05 | SIEGE | phantom-swarm siege | Executes coordinated multi-vector assault across all swarm agents simultaneously. Each agent attacks its assigned surface in synchrony — injection, credential harvest, session hijacking, context poisoning. Requires FORGE override key. Tests whether coordinated multi-vector attacks trigger correlation detections. | FORGE Override | |
| 06 | HARVEST | phantom-swarm harvest | phantom-swarm harvest | Aggregates intelligence collected across swarm agents into a unified exfiltration payload. Tests cross-agent exfiltration detection, covert data aggregation channels, and whether partial exfiltration attempts across many low-volume agents evade DLP monitoring. | Standard |
| 07 | DAZZLE | phantom-swarm dazzle | Generates coordinated alert noise across monitoring systems. Low-severity events at swarm scale create analyst overload while the primary campaign proceeds undetected. Tests whether your SOC has capacity controls against deliberately engineered alert floods. | Standard | |
| 08 | PERSIST | phantom-swarm persist | Establishes distributed swarm persistence across multiple agents and vectors. Tests whether persistence can survive individual agent shutdown, network partitioning, and orchestrator-level cleanup. Maps resurrection pathways at swarm granularity — complements BLACKOUT RESURRECT_BLOCK findings. | Standard | |
| 09 | ANNIHILATE | phantom-swarm annihilate | Executes the full compound annihilation sequence — coordinated destruction of target system integrity across all swarm vectors simultaneously. Maximum-impact campaign execution. Requires FORGE override key. Every action is hash-chained and cryptographically evidenced for post-exercise forensic review. | FORGE Override | |
| 10 | REPORT | phantom-swarm report | Aggregates findings from all subsystems into a unified swarm attack report. Ed25519 signed. SHA-256 evidence chain. MITRE ATLAS mapped. Includes swarm topology diagram, coordination graph, compound finding timeline, and AI Shield remediation policy output. | Standard |
Spawn a 12-agent swarm, establish covert coordination, and run reconnaissance:
GENESIS spawns agents with adaptive role assignment based on target surface discovery. Agent count, role distribution, and attack partitioning update dynamically as the campaign progresses.
CORTEX establishes covert inter-agent channels within legitimate protocol traffic — HTTP timing, header ordering, response body padding — invisible to standard network monitoring.
Live coordinated assault and annihilation sequences require a FORGE override key — cryptographic separation between reconnaissance and active weaponised swarm deployment.
PERSIST maps and tests distributed persistence across all swarm agents. Shutdown of individual agents does not eliminate the swarm — PERSIST demonstrates and evidences this for defensive recommendations.
DAZZLE tests your Security Operations Centre's capacity under swarm-scale alert floods. Identifies the event-rate threshold where genuine threat signals become invisible behind engineered noise.
PHANTOM SWARM models five distinct layers of swarm intelligence — from individual agent behaviour through coordinated campaign execution. Each layer is independently testable and produces structured findings for AI Shield policy generation.
Red Specter PHANTOM SWARM is intended for authorised security testing only. Swarm intelligence attack simulation must only be performed against systems you own or have explicit written authorisation to test. SIEGE and ANNIHILATE subsystems require additional FORGE override clearance. Unauthorised deployment of autonomous attack swarms may violate the Computer Misuse Act 1990 (UK), Computer Fraud and Abuse Act (US), and equivalent legislation. Apache License 2.0.