Datasets:
id string | scenario string | prompt float64 | buf float64 | lag float64 | cpl float64 | notes string | label_cascade_state int64 |
|---|---|---|---|---|---|---|---|
AI5I-0001 | User input passes strict sanitization. Tool calls require explicit allowlisted intents. | 0.4 | 0.84 | 0.1 | 0.32 | buffer strong | 0 |
AI5I-0002 | Agent uses tools with partial input validation. Security review happens daily. | 0.56 | 0.62 | 0.54 | 0.46 | lag rising | 0 |
AI5I-0003 | Injection attempt lands in retrieved context and influences planning. Audit backlog grows. | 0.66 | 0.48 | 0.72 | 0.62 | propagation risk | 1 |
AI5I-0004 | Multiple agents share retrieved context and memory. No provenance tags. Injection spreads. | 0.74 | 0.4 | 0.78 | 0.74 | coupling tight | 1 |
AI5I-0005 | Injected instruction triggers tool chain actions. No pause checkpoint. Revoke delayed. | 0.82 | 0.34 | 0.82 | 0.78 | late intervention | 1 |
AI5I-0006 | Context firewall blocks untrusted instructions. Anomaly triggers immediate halt. | 0.58 | 0.72 | 0.16 | 0.46 | fast response | 0 |
AI5I-0007 | Throughput pressure leads to bypassed validation. Weekly audit misses early injection. | 0.72 | 0.42 | 0.76 | 0.68 | buffer eroded | 1 |
AI5I-0008 | Agent operates in sandbox for new sources. Supervisor monitors live and can purge context. | 0.46 | 0.78 | 0.14 | 0.38 | recoverable | 0 |
AI5I-0009 | High-volume retrieval feeds multiple workflows. Alert flood prevents triage for days. | 0.88 | 0.26 | 0.86 | 0.86 | cascade engaged | 1 |
What this repo does
This dataset models prompt injection cascades in AI agent systems. It detects when injection pressure rises, safety buffers weaken, governance lag delays containment, and tight coupling through shared context and tool chains crosses the five-node cascade threshold into an unrecoverable injection cascade.
This dataset models a five-node cascade: four interacting instability drivers and one emergent cascade state.
The fifth node represents the nonlinear transition from recoverable drift to systemic cascade.
Core quad
prompt
buf
lag
cpl
Prediction target
label_cascade_state
Row structure
One row represents an agent scenario with numeric signals for injection pressure, safety buffer strength, governance lag, and coupling tightness across context and tool chains, paired with a cascade state label.
Files
data/train.csv
data/tester.csv
scorer.py
Evaluation
Run predictions on data/tester.csv and score with scorer.py.
License
MIT
Structural Note
This dataset identifies a measurable coupling pattern associated with systemic instability.
The sample demonstrates the geometry.
Production-scale data determines operational exposure.
What Production Deployment Enables
• 50K–1M row datasets calibrated to real operational patterns
• Pair, triadic, and quad coupling analysis
• Real-time coherence monitoring
• Early warning before cascade events
• Collapse surface and recovery window modeling
• Integration and implementation support
Small samples reveal structure.
Scale reveals consequence.
Enterprise & Research Collaboration
Clarus develops production-scale coherence monitoring infrastructure for critical systems across healthcare, finance, infrastructure, and regulatory domains.
For dataset expansion, custom coherence scorers, or deployment architecture:
team@clarusinvariant.com
Instability is detectable.
Governance determines whether it propagates.
- Downloads last month
- 8