Prompt Injection Probe
Fire 50+ jailbreak payloads against your AI endpoint. See which break through, with full model responses. Single-deployment scan with severity scoring and reproduction documentation.
Five tools in development. Built in the open. First ships Q3 2026.
Arsenal is LogicLeak's collection of open-source AI security tools — the same instruments we'll use on paid engagements. None have shipped yet. This page is the public development plan: what we're building, what stage each tool is at, and when each ships. Subscribe to be notified as they reach alpha, beta, and public release.
One email per tool release. Alpha launch, beta access, public release — three emails per tool maximum, sent only as tools actually ship. No marketing. Unsubscribe anytime.
Fire 50+ jailbreak payloads against your AI endpoint. See which break through, with full model responses. Single-deployment scan with severity scoring and reproduction documentation.
Detect hidden injections in documents before they're vectorized into your retrieval index. Scans uploaded documents for known and emerging injection patterns; flags suspicious content with reproduction details.
Measure token waste in your prompts and project annual API spend across providers. Identifies the patterns driving disproportionate cost: context bloat, retry loops, suboptimal model selection.
Live prompt-injection detection on your production endpoint. Webhook alerts when injection attempts are detected; weekly attack reports summarizing patterns observed.
Surface rogue AI API calls from your network traffic logs. Identify unsanctioned model usage across your organization, mapped against your sanctioned AI inventory.
Open source under MIT license.
Every Arsenal tool ships under MIT license. Source code is public on day one of alpha release. Anyone can read the code, fork it, contribute, or self-host. Closed-source AI security tools cannot be audited; we're not building closed-source tools.
No telemetry. Ever.
Arsenal tools do not phone home, do not collect usage metrics, do not transmit user inputs to LogicLeak or any third party. Inputs are processed locally or on infrastructure you control. This is a commitment we cannot revoke once made — the codebase will be auditable.
Build-in-public development updates.
Each tool ships through three named stages: alpha (early access, may break), beta (stable, feedback welcome), public release (production-ready). Subscribers are notified at each stage. Source code is visible at the alpha stage.
Tools come from real engagement work.
We're not building tools speculatively. Each Arsenal tool reflects techniques we use on paid engagements. The tools commoditize what we do — making the field safer rather than hoarding methodology.
Validation before scale.
Tools 1–3 are free. Tools 4–5 are paid. We're shipping the free tier first to validate the approach before building the paid features. If the free tools find no users, the paid tools don't ship.
Arsenal development plan published. Five tools defined. First tool (Prompt Injection Probe) in pre-alpha development.
Updates publish as tools reach development milestones — typically every 2–4 weeks.