v1.23.0 v1.7.20
AIPentest/CyberStrikeAISep 22, 2026 → Sep 22, 20261 day between releases
What's New
Replaced PDF text extraction with a TypeSafe Jev HITL audit backend for enhanced security.
Breaking Changes
- PDF text extraction features removed and replaced by TypeSafe Jev HITL audit backend.
- Chat model JSON output dependency removed; switch to structured pass/block scoring.
New Features
- TypeSafe Jev human-in-the-loop approval engine integration.
- Structured pass/block scoring for tool calls.
- Parallel evaluation of destructive risk and penetration intent.
- Default blocking of irreversible data loss and critical config changes.
- Custom policy overlay support for organizational rules.
Improvements
- Enhanced safety and control via structured HITL audit backend.
- Replaced chat model JSON with parallel risk and intent judgment.
- Default policies block live-system destruction and data loss.
- Custom organizational policies can be layered over default rules.
- OpenAI-compatible reviewer remains available for legacy use.
Upgrade Notes
Remove dependency on chat model JSON outputs and switch to structured pass/block scoring via TypeSafe Jev. Configure default rules to block destructive actions like data loss, and layer custom policies as overlays.
The Two Releases
1.23.0 — Text-layer repairs
Sep 22, 2026· 30 days ago
Version 1.23.0 focuses on text-layer repairs, including handling large Form XObject bounding boxes, fixing right-to-left text reading order, and improving ligature glyph name decoding.
Full release notesSep 22, 2026· 29 days ago
CyberStrikeAI integrates TypeSafe Jev as a new Human-in-the-Loop (HITL) audit backend to enhance safety and control. The system replaces reliance on chat model JSON outputs with structured pass/block scoring for tool calls. It establishes default rules to approve authorized penetration while blocking destructive actions like data loss and configuration changes.
Full release notes