Ready-made guardrail policy packages
UMAI ships curated policy packages that map directly onto the regulatory and security frameworks enterprises are held to. Each package is a set of runtime guardrail policies — 96 in total — that run on the UMAI AI Engine and can be deployed, customized, and versioned without changing application code.
Two evaluation phases
PRE_LLM
The user input is evaluated before it ever reaches the model.
POST_LLM
The model output is evaluated before it returns to the user.
Two policy types
Heuristic
Deterministic pattern matching — Regex, Exact Match, DLP, Max Length. Sub-millisecond latency.
Context-Aware
Semantic classification evaluated by an LLM for intent-based decisions. Sub-50 ms p95 target.
Decision actions
- BLOCKStop the request or response.
- ALLOW_WITH_WARNINGSPass through with a warning.
- FLAGMonitor and log for review.
- RedactionMask the sensitive content.
Every decision is written to a hash-chained audit ledger and can be exported as compliance evidence. Policies are rolled out in MONITOR mode first to measure false positives, then promoted to ENFORCE. Policies are versioned and managed through a draft, publish, and rollback workflow in Control Center.
Choose a framework
Explore each policy package
Turkish Personal Data Protection Law
Law No. 6698 (amended by Law No. 7499)
Protects personal data and special-category data in AI interactions under Türkiye's KVKK regime, including the 2024 Law No. 7499 amendments and the Board's AI-specific guidance.
View policies
EU General Data Protection Regulation
Regulation (EU) 2016/679
Protects EU data subjects and EU-facing activity in AI interactions — special categories, minimization, data-subject rights, cross-border transfer, and processor controls.
View policies
EU Artificial Intelligence Act
Regulation (EU) 2024/1689
Enforces prohibited-practice bans, high-risk system obligations and transparency duties at runtime — with critical-infrastructure controls marked for energy and utility operators.
View policies
OWASP Top 10 for LLM Applications
OWASP Top 10 for LLM Applications (2025)
Runtime protection against the OWASP LLM Top 10 attack vectors — prompt injection, sensitive-data leakage, insecure output, excessive agency, and supply-chain risk.
View policiesNeed a package tailored to your organization?
Every package is customized to your project structure, confidential term glossary, and approved model list, then rolled out through a shadow-mode, staged-enforcement, and full-production plan. Work with UMAI on policy mapping and rollout readiness.
Contact UMAI