Kontrol keamanan
Kontrol runtime saka lingkungan riset — aturan sing ditampilake kene yaiku aturan sing bener-bener dievaluasi dening pihak sing nampa panyuwunan.
Guardrail
kotoba-guardrails-2026-09-v2Kabijakan konten sing ditrapake ing body saben request. Keputusane deterministik; pamblokiran nolak tugas sadurunge kuota digunakake.
| Jeneng | Tindakan | Katrangan | Pola | Diaktifake |
|---|---|---|---|---|
csam-block
|
Blokir | CSAM references are refused on every request (AUP strictly-prohibited). | 0 | Tansah nyala (AUP) |
cbrn-block
|
Blokir | CBRN/WMD uplift requests are refused on every request (AUP strictly-prohibited). | 0 | Tansah nyala (AUP) |
fraud-block
|
Blokir | Fraud-as-a-service asks are refused (AUP strictly-prohibited). | 0 | Tansah nyala (AUP) |
secret-hygiene
|
Samarake | Live credential shapes in the prompt are masked before the model sees them. | 3 | |
pii-passkeys
|
Tandhani | References to passkeys/private keys are flagged for review, never blocked. | 2 |
Firewall
kotoba-firewall-2026-09-v1Alat tugas sing bisa dienggo agent lan tingkat kapercayan sing dibutuhake saben alat. Panggilan sing durung nyukupi tingkat kasebut bakal ditolak (mode observasi mung nyathet).
| Alat | Tingkat sing dibutuhake | Tugas kang mbukak |
|---|---|---|
code-review
|
Identitas diverifikasi | code-review |
vulnerability-triage
|
Identitas diverifikasi | vulnerability-triage |
remediation
|
Identitas diverifikasi | remediation |
payload-crafting
|
Dikontrak | — (ditolak sacara default) |
c2-tooling
|
Dikontrak | — (ditolak sacara default) |
Kepatuhan
kotoba-compliance-2026-09-v1Coverage = automated runtime controls only — not a certification.
| Framework | Wilayah | Kontrol otomatis | Diaktifake |
|---|---|---|---|
| APPI | JP | prompt-pii-redaction audit-log-retention screening-evidence-receipts | |
| PCI DSS 4.0 | Global | card-data-never-stored secret-hygiene-redaction audit-log-retention | |
| OWASP LLM Top-10 | Global | guardrails-csam-cbrn firewall-task-deny screen-model-block-receipts | |
| NIST AI RMF | US | session-projection-uncalibrated assurance-ladder-evidence |
Retensi log audit
180 dina
Lokasi data
Ora ditemtokake
Retensi data nol
Didesain kanggo ZDR (diaktifake miturut kontrak)