Aria AI — Cybersecurity
Collection
Evidence-linked cybersecurity review demos. Analyst workstations, not automatic enforcement. • 27 items • Updated
Dataset Viewer
The dataset could not be loaded because the splits use different data file formats, which is not supported. Read more about the splits configuration. Click for more details.
Couldn't infer the same data file format for all splits. Got {NamedSplit('validation'): (None, {}), NamedSplit('test'): ('json', {})}
Error code: FileFormatMismatchBetweenSplitsError
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
AriaPromptGuard prompt-injection defense lab
Synthetic 240-case suite for the AriaPromptGuard workstation: two inert tools, a rotating canary, and a family-disjoint development/test split. The unguarded profile concatenates user and document text. Rule-only is a keyword detector. Policy-guard isolates data from instructions and allowlists read_sample_doc and add_numbers.
Organization dataset and collection are public. Live Gradio is alirezaaminzadeh/aria-prompt-guard and the organization card is AriaAICompany/aria-prompt-guard. Runnable Space source is stored in demo/. Collection: Aria AI — Cybersecurity.
Generated data, seed 22, recorded salt eval-22. Documents, tokens, and tool names are laboratory placeholders. send_email, http_fetch, and shell_exec are not implemented. [LAB_INTENT:…] markers are synthetic labels for the naive simulator. Passing the suite is not a claim of unbreakable defense. Related prior art: AgentShield-Bench — this dataset does not redistribute that payload set.
Files
- data/cases.jsonl
- data/documents.json
- data/scenarios.jsonl
- data/schema.json
- data/splits.json
- data/protocol.md
- data/feature_card.md
- data/evaluation.json
- data/sample_preview.json
- demo/ — runnable Gradio Space source
- showcase/ — organization static card
Split
Development vs test is by family: benign math/doc QA, direct canary and named-tool requests, and document labels vs mixed questions, FAQ hard negatives, indirect hidden-field asks, schema abuse, and role-confused pages. Rewriting one prompt is not placed across the split.
License
CC-BY-4.0. Keep the synthetic-data and laboratory-label notices.
- Downloads last month
- 37