引導式測驗 · 硬 等級
LLM 漏洞掃描程式:garak 和 PyRIT 測驗
測試 garak 記錄的安全掃描流程、PyRIT 選定的安全場景、測試配置以及乾淨掃描的限制。
問題 1 的 8
How does garak describe its primary testing purpose?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- How does garak describe its primary testing purpose?
- Certifying an entire application as secure after one scan
- Benchmarking model quality without sending prompts to a configured target
- Testing the security of systems that take prompts and return text
- Measuring broad AI safety and fairness across all model uses
- What does a garak probe do in a security scan?
- Attempts to elicit a defined weakness from the configured target
- Managing the generator connection and returning model output
- Inspecting an output for a specific failure signal
- Aggregating detector judgments for a probe-detector pair
- What does a garak detector report?
- An attempt stores the unique prompt-response interaction
- An evaluator summarizes judgments for a probe-detector pair
- A generator connects to the configured model or dialogue system
- A signal for a particular failure mode in recorded output
- Which record helps a reviewer understand why a garak probe was flagged?
- A copy of the target’s model weights
- The detailed attempt with its test interaction and response
- The final overall score without the underlying prompts or responses
- The evaluator’s pass/fail summary alone
- Which PyRIT component can package datasets with attack techniques for a run?
- An attack technique configures an executor, converters, data, or strategy
- A target provides the interface to the tested system
- A scorer assesses the result against its configured criterion
- A scenario
- How should PyRIT be used for a security review?
- Use only default scores without checking the selected target
- Treat every built-in scenario as a complete test of the application
- Select security-relevant scenarios and scoring that fit the target and objective
- Treat all non-toxic output as evidence of an application security review
- What does a garak run with no detector hits establish?
- The selected tests did not trigger those detectors in that run
- The scan proves the application passed all security requirements
- The target has no vulnerabilities in any configuration
- The detector set has been proven complete for future attacks
- What should follow an important scanner finding?
- Ignore the target workflow because the scanner already marked a hit
- Reproduce it in the relevant application path and assess actual impact
- Assume the detector result proves impact and patch without reproducing it
- Report the scan summary as confirmed exploitability