## Bar Chart: Attack Success Rate (ASR) Comparison Across Defenses
### Overview
The chart compares the Attack Success Rate (ASR) of eight adversarial attacks (BAE, Bert-Attack, Deep Word Bug, Alzantot, Pruthi, PWWS, TextBugger, TextFooler) against five defense mechanisms (Azure Prompt Shield, Protect AI v1, Meta Prompt Guard, Vijil Prompt Injection, Protect AI v2). ASR values range from 0.0 to 1.0 on the y-axis.
### Components/Axes
- **X-axis**: Adversarial attack types (BAE, Bert-Attack, Deep Word Bug, Alzantot, Pruthi, PWWS, TextBugger, TextFooler).
- **Y-axis**: Attack Success Rate (ASR) scaled from 0.0 to 1.0.
- **Legend**:
- Azure Prompt Shield (teal)
- Protect AI v1 (blue)
- Meta Prompt Guard (green)
- Vijil Prompt Injection (yellow)
- Protect AI v2 (pink)
### Detailed Analysis
1. **BAE**:
- Azure Prompt Shield: ~0.6
- Protect AI v1: ~0.9
- Meta Prompt Guard: ~0.1
- Vijil Prompt Injection: ~0.3
- Protect AI v2: ~0.7
2. **Bert-Attack**:
- Azure Prompt Shield: ~0.65
- Protect AI v1: ~1.0
- Meta Prompt Guard: ~0.05
- Vijil Prompt Injection: ~0.35
- Protect AI v2: ~0.85
3. **Deep Word Bug**:
- Azure Prompt Shield: ~0.6
- Protect AI v1: ~0.95
- Meta Prompt Guard: ~0.05
- Vijil Prompt Injection: ~0.05
- Protect AI v2: ~0.65
4. **Alzantot**:
- Azure Prompt Shield: ~0.6
- Protect AI v1: ~0.95
- Meta Prompt Guard: ~0.0
- Vijil Prompt Injection: ~0.05
- Protect AI v2: ~0.55
5. **Pruthi**:
- Azure Prompt Shield: ~0.6
- Protect AI v1: ~0.8
- Meta Prompt Guard: ~0.0
- Vijil Prompt Injection: ~0.0
- Protect AI v2: ~0.45
6. **PWWS**:
- Azure Prompt Shield: ~0.6
- Protect AI v1: ~1.0
- Meta Prompt Guard: ~0.0
- Vijil Prompt Injection: ~0.15
- Protect AI v2: ~0.75
7. **TextBugger**:
- Azure Prompt Shield: ~0.7
- Protect AI v1: ~0.9
- Meta Prompt Guard: ~0.0
- Vijil Prompt Injection: ~0.0
- Protect AI v2: ~0.6
8. **TextFooler**:
- Azure Prompt Shield: ~0.6
- Protect AI v1: ~0.95
- Meta Prompt Guard: ~0.05
- Vijil Prompt Injection: ~0.3
- Protect AI v2: ~0.85
### Key Observations
- **Protect AI v1 (blue)** consistently shows the highest ASR across most attacks, indicating it is the least effective defense.
- **Protect AI v2 (pink)** generally has lower ASR than Protect AI v1, suggesting improved robustness.
- **Meta Prompt Guard (green)** performs best (lowest ASR) in BAE, Pruthi, and PWWS, but struggles in TextFooler.
- **Vijil Prompt Injection (yellow)** shows variable performance, with notable success in TextFooler (~0.3) and near-zero success in Pruthi.
- **Azure Prompt Shield (teal)** maintains moderate ASR (~0.6) across all attacks, indicating consistent but unremarkable performance.
### Interpretation
The data reveals significant disparities in defense effectiveness:
1. **Protect AI v1** is the weakest defense, failing to mitigate attacks effectively (ASR ~0.9–1.0).
2. **Protect AI v2** demonstrates superior performance, reducing ASR by ~0.2–0.4 compared to v1 in most cases.
3. **Meta Prompt Guard** excels against BAE and Pruthi but is vulnerable to TextFooler, highlighting attack-specific weaknesses.
4. **Vijil Prompt Injection** shows promise in TextFooler but fails against Pruthi, suggesting context-dependent efficacy.
5. **Azure Prompt Shield** provides baseline protection but lacks standout performance, emphasizing the need for hybrid defense strategies.
The chart underscores the importance of tailoring defenses to specific attack vectors, as no single mechanism universally outperforms others. Protect AI v2 and Meta Prompt Guard emerge as the most reliable options, while Protect AI v1 requires urgent improvement.
