無限
// Adversarial rare-event assurance

Your software has never met its worst day. We have.

Mugen.Codes is a one-stop shop for auditing and hardening software in defense, space and allied industries against the rare, deliberate attacks that test suites never cover: GNSS spoofing, jamming, timing manipulation, sensor injection and more. Our engine, PHINN-SEC, synthesises thousands of plausible attacks from a handful of real ones.

GNSS_PVT / position error (m) · SIL harness SYNTH ATTACK #0412
nominal receiver PHINN-SEC spoof variants detector threshold
Variant familySlow-drag spoof
Seed events3 real
Synth. variants1,000
01 — The problem

Attacks live in the tail.
Test suites live in the middle.

Adversarial events against navigation, timing and sensing are rare, deliberately novel and usually classified. You can't regression-test against an attack you have only seen three times, and the next one will be shaped to slip past whatever caught the last.

where your tests are where attacks are nominal ← → adversarial
Schematic, not to scale. Three red points = the only real-world samples most programmes ever get.

Scarce data

A spoofing incident or a jamming campaign yields a handful of traces, often under classification. Nothing like enough to train a detector or prove a fallback.

Adaptive adversaries

Attackers iterate. The next spoof will ramp slower, hide under noise or arrive mid-handover. Signature-based testing fights the last war.

Siloed tools

Static analysis, pen-testing, RF simulation and HIL rigs are bought from different vendors and never meet. Nobody owns the attack end to end.

Mugen closes the loop

One team audits the code, synthesises the rare attacks, injects them into your harness, ships the fixes and hands over the evidence.

02 — The engine

PHINN-SEC: PHINN, pointed at your attack surface.

PHINN was built to synthesise rare-event time series, such as extreme space-weather storms, from their topological fingerprints (Betti curves). PHINN-SEC applies the same mathematics to adversarial events: it learns the shape of an attack from a few real examples and generates thousands of new, physically plausible variants your software has never seen.

01

Fingerprint

Extract the topological signature of each known attack from signal, telemetry and log traces.

Betti curves · persistence
02

Synthesise

Generate thousands of novel variants conditioned on that signature: same shape of threat, new timing, amplitude and stealth.

rare-event generation
03

Inject

Replay variants through your software-in-the-loop, HIL or digital twin. Measure what breaks, and when.

SIL · HIL · twin
04

Harden

Our engineers ship the fixes: integrity checks, detectors, sensor-fusion guards, safe-mode logic, code patches.

code + detectors
05

Prove

Re-run the full synthetic campaign. Deliver an evidence pack mapped to your assurance framework.

evidence package

PHINN · Space weather

  • A few Carrington-class storms on record
  • Synthesise thousands of plausible extreme storms
  • Stress satellite portfolios, catalogues and insurance books
  • Quantify tail exposure before it happens

PHINN-SEC · Adversarial events

  • A few observed spoofing or jamming incidents
  • Synthesise thousands of novel, stealthier attack variants
  • Stress navigation, timing, sensing and control software
  • Find and fix the failure before an adversary does
03 — Threat Lab

See a synthetic campaign.

Pick a threat family. Green is the nominal trace; red lines are PHINN-SEC variants generated around the same fingerprint. Hit Re-synthesise to draw a fresh batch.

—
What it is

—

—

What we harden

    Illustrative visualisation for explanation only. Real campaigns run on your data, in your environment, under your security controls.

    04 — Threat catalogue

    The rare events we audit for.

    If it is rare, deliberate and hits software through a signal, a sensor, a clock or a link, it belongs here.

    T-NAV-01

    GNSS spoofing

    Counterfeit signals that drag position, velocity or time slowly enough to evade checks.

    UAS · maritime · aviation · munitions
    T-NAV-02

    Jamming

    Wideband, swept and pulsed denial that forces fallback to inertial or dead reckoning.

    all PNT users
    T-NAV-03

    Meaconing

    Delayed rebroadcast of authentic signals that defeats naive authentication.

    aviation · maritime
    T-TIME-01

    Timing attacks

    Manipulated PTP/NTP or GNSS time feeding grids, telecom and trading systems.

    grid · 5G · data centres
    T-SENS-01

    Sensor false-data injection

    Crafted IMU, radar, lidar or EO/IR inputs that steer fusion and autonomy stacks.

    autonomy · C-UAS · ISR
    T-LINK-01

    Datalink replay & injection

    Replayed or forged frames on C2, telemetry and tactical links.

    UAS · satellites · ground stations
    T-MAR-01

    AIS / ADS-B spoofing

    Ghost vessels and aircraft, identity swaps and track manipulation that pollute the picture.

    MDA · ATM · ports
    T-ML-01

    ML evasion

    Adversarial inputs that blind classifiers in target recognition and anomaly detection.

    ISR · autonomy · cyber
    05 — One-stop shop

    Audit. Harden. Build. Watch.

    Four services, one team, one evidence trail. Engage for a single audit or let us own adversarial assurance across the programme lifecycle.

    AUDIT

    Adversarial rare-event audit

    A structured review of how your software behaves when signals, sensors and clocks lie.

    • Threat model & attack-surface map
    • Code & architecture review
    • PHINN-SEC campaign on your traces
    • Failure-mode ranking
    Output → Audit report + reproducible failing cases
    HARDEN

    Remediation engineering

    We don't hand you a list and leave. Our engineers write and verify the fixes inside your codebase.

    • Integrity & consistency monitors
    • Multi-sensor cross-checks
    • Graceful degradation & safe modes
    • Detector tuning on synthetic tails
    Output → Merged patches + before/after campaign results
    BUILD

    Secure-by-design development

    New navigation, timing, autonomy or ground-segment software, built with the attack campaign in CI from day one.

    • Resilient PNT & sensor-fusion modules
    • Hardened C2 & telemetry links
    • Synthetic attacks in the CI pipeline
    • Assurance artefacts as you build
    Output → Production code with continuous evidence
    WATCH

    Continuous assurance

    Threats evolve. When a new incident is observed anywhere, we fingerprint it, synthesise its family and re-test your releases.

    • New-threat fingerprinting
    • Release-gate attack campaigns
    • Regression on every build
    • Quarterly assurance briefings
    Output → Rolling evidence + threat bulletins
    06 — Where we work

    Defense, space and the industries that depend on them.

    IND-01

    Defense & uncrewed systems

    UAS, loitering munitions, C-UAS, ISR and mission software operating in contested EW environments.

    GNSS denial · link injection · ML evasion
    IND-02

    Space & ground segment

    Flight software, TT&C, ground stations and constellation operations.

    command-link intrusion · telemetry spoofing
    IND-03

    Maritime

    Navigation bridges, maritime domain awareness, ports and autonomous vessels.

    AIS spoofing · GNSS drag · ghost tracks
    IND-04

    Aviation & air traffic

    Avionics, ATM surveillance and airport systems.

    ADS-B injection · meaconing
    IND-05

    Energy & grid

    Substations, synchrophasors and protection systems that run on precise time.

    PTP / GNSS time manipulation
    IND-06

    Telecom, finance & heavy industry

    5G timing, timestamped trading and industrial control in steel, mining and logistics.

    time-source attacks · sensor injection
    Evidence packs aligned to → NIST SP 800-53IEC 62443DO-178C / DO-326AMIL-STD-882EISO/SAE 21434CCSDS securityNIST PNT profile
    07 — How an engagement runs

    From first call to hardened release.

    STEP 1

    Scope

    Threat-model workshop. We pick the systems, attack families and harness to target.

    → statement of work
    STEP 2

    Audit sprint

    Code review plus a first PHINN-SEC campaign seeded from your traces or ours.

    → failure-mode report
    STEP 3

    Harden

    Our engineers remediate alongside yours, verifying each fix against the synthetic tail.

    → patches + results
    STEP 4

    Prove & watch

    Full re-run, evidence pack, and optional continuous assurance on future releases.

    → evidence package
    08 — Built on published research

    The maths is public. The engagements are not.

    PHINN's rare-event synthesis method is published openly so programme offices, primes and insurers can scrutinise it. Mugen.Codes is the engineering arm that turns that method into hardened software.

    arXiv:2606.15452

    PHINN: rare-event time-series synthesis

    Topological fingerprints (Betti curves) condition the generation of rare-event time series, producing plausible extremes far beyond the observed record. First applied to space weather; PHINN-SEC extends it to adversarial events against software.

    Read the paper →
    無限
    Mugen (無限) — infinite

    Test against infinite attacks, not the three you've seen.

    Tell us the system and the threat that keeps you up at night. We'll come back with a scoped audit plan within a week.

    ceo@capa.cloud →