CORROBORATE
CORROBORATE
Try Live PoCInspector
ADVERSARY BENCHMARK GENERATOR

CHIMERA Attack Workbench.

Synthesizing composite "Frankenstein" sessions. By splicing genuine human fragments at pause boundaries (velocity ≈ 0), attackers bypass single-channel ML while corrupting cross-modal synchrony.

Adversarial Synthesis Pipeline

How CHIMERA Synthesizes Bypasses

STEP 01

Donor Selection

Selects 3 distinct human sessions to prevent single-source trajectory fingerprinting.

STEP 02

Pause Segmentation

Cuts mouse paths at stationary dwell points (v ≈ 0) to avoid jerk spikes.

STEP 03

Spatial Retarget

Translates and scales coordinates to match target input form bounding boxes.

STEP 04

Micro-Jitter

Applies 0.90x–1.15x retiming to defeat historical duplicate matching.

STEP 05

Keystroke Overlay

Grafts an independent foreign typing cadence onto the spliced timeline.

Multi-Donor Splicer

Configure Attack Parameters

Target Attack: A6_chimera

Initial ballistic approach toward form field.

Spliced at velocity ≈ 0 pause dwell boundary.

Independent typing cadence overlaid on timeline.

Multi-Channel Spliced Composition Preview
Channel A: Mouse TrajectoryCut at t = 1200ms (v ≈ 0)
Donor 1 (s-0001)
Donor 2 (s-0002)
Channel B: Keystroke RhythmForeign typing cadence overlay
Overlaid Key Cadence from Donor 3 (s-0003)
Micro-Jitter & Time-Scale Factor: 1.05x0.90x – 1.15x
Literature Foundation

Why Spliced Attacks Bypass Single-Channel ML

01

Individual Fragments are 100% Genuine

Because every donor fragment came from a real human, intra-channel metrics (micro-tremor, speed distributions) pass statistical normality checks.

02

Cuts Occur at Natural Pauses

CHIMERA detects moments where mouse velocity is zero ($v \approx 0$). Splicing here generates zero acceleration or jerk spikes.

03

The Fatal Cross-Modal Desync

While each channel looks human in isolation, their cross-modal timing violates Card-Moran-Newell KLM single-attentional bounds, which CORROBORATE catches.