"48.9% of organizations are entirely blind to AI agent behavior" · Salt Security, April 2026

ORILink · changelog

Changelog

ORILink is actively developed. This log tracks architecture updates, new enforcement layers, and SDK changes. All updates are included in your license. No action required on your end.

June 16, 2026

v1.0.3

  • Gate 1.5 M3 semantic intent monitoring layer: 142-pattern corpus, 80% taint signal accuracy, 0% false positives
  • Priority 2: Provenance Envelope v2. HMAC-SHA256 field sealing, SHA-256 content binding, immutable origin lock, trust ratcheting
  • Priority 3: Inter-gate taint propagation. Low-trust inbound content automatically increases scrutiny on derived outputs
  • MCP hardening layer: 36/36 tests, 3.14ms latency
  • Individual SDK test cases: 540 → 602
  • Total combined test cases: 1,310
  • Total stack latency updated: ~1ms → ~13ms (12ms is M3 semantic layer, a deliberate trade for broader attack coverage)
  • Production dashboard: real-time enforcement data pipeline stabilized
June 6, 2026

v1.0.2

  • Windows installer: ORILink-1.0.0-Setup.exe (built, tested)
  • Linux installer: orilink_1.0.0_amd64.deb (built, tested, systemd working)
  • npm packages: orilink-individual-1.0.0.tgz
  • Platform selection flow: before payment, not after
  • Individual license activations: effectively unlimited (999)
  • Business SDK test cases: 488
  • DMARC record added
June 2, 2026

v1.0.1

  • addToAllowlist() locked at runtime: immutable after init across all gate layers
  • License key system live: validate-license Edge Function deployed
  • Purchase confirmation emails live
May 2026

v1.0.0

  • Gate 1 inbound trust annotation: 100% block rate, 0% false positives
  • Gate 2 outbound intent enforcement: four-layer classifier
  • ORIMesh A2A trust enforcement
  • ORIMark audit registry
  • Validated across 6 LLM architectures (Llama 3, Mistral, Gemma 2, GPT-4o, GPT-4o mini, Claude)
  • Model-agnostic proof confirmed across full hardening spectrum