引導式測驗 · 中等 等級
數位鑑識AI測驗
測試人工智慧數位鑑識:合法範圍、獲取和保存、工具驗證、再現性、模型錯誤、雜湊值和專家報告。
問題 1 的 8
An AI classifier flags a phone image as relevant. What should the examiner conclude immediately?
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。
- An AI classifier flags a phone image as relevant. What should the examiner conclude immediately?
- That the image is authentic and admissible
- Only that the item meets the tool’s flagging criteria and needs examination
- That the image is within the warrant scope
- That the person in the image committed an offense
- What does a matching cryptographic hash support when comparing two acquired copies?
- That the device owner created every file
- That the acquisition stayed within the warrant
- That the evidence proves the suspected event
- That their bytes match the recorded digest, not that the original data were lawful or genuine
- Why must AI triage remain within a warrant’s scope?
- Automated search does not expand the legal authority that limits the examination
- A model’s confidence creates new search authority
- Triage is exempt from warrant limits
- Only human searches can exceed scope
- Which tool-validation practice is strongest for a forensic task?
- Rely only on the vendor’s general accuracy claim
- Run the tool once on the case evidence
- Test known inputs and measure errors for the intended artifact type
- Use a benchmark for an unrelated data type
- A language model summarizes messages but omits a negation. How should the examiner use the summary?
- Quote it as an exact transcript
- Treat it as a navigation aid and verify statements in the original messages
- Discard the source extraction after summarizing
- Use the summary’s confidence to restore the missing word
- Which information makes an AI-assisted examination more reproducible?
- Tool and version, settings, preprocessing, thresholds, source data, and analyst actions
- Only the final conclusion
- The vendor name without configuration details
- A confidence score without the tested files
- Why preserve an original acquisition as read-only and work from verified copies?
- To guarantee the data are authentic
- To eliminate the need for legal authority
- To ensure a model cannot produce false results
- To protect the source and detect later changes during analysis
- How should a forensic report present model outputs?
- State model output as fact to avoid confusing the court
- Omit the model version if results look obvious
- Separate observed artifacts from inferences and explain validation, limitations, and uncertainty
- Report only items the model marked with high confidence