The latest news, updates, and insights from our team on our mission to build AI systems that are safe, transparent, and aligned with human values.
6 posts tagged Guardrails
When OpenAI, Anthropic, Google, or others release a new LLM, we're curious whether it makes sense to use it as a safety guardrail. Trying a few prompts out usually gives us a first impression; however, a proper comparison with other models takes more work. We built unsafe-eval to make those comparisons easier to run, and we're now sharing it.
Sometimes your AI assistant should use a reply you've already written, or leave a request to someone on your team. Following feedback from our users, we built ScopeGuard-2 to recognize when each is needed. We're releasing it with more control over what each prediction returns, and new models that match frontier models on accuracy in our evaluations while responding much faster.
We're introducing the unsafe-eval leaderboard to help compare models for AI safety tasks. Our first results examine frontier models and dedicated classifiers as safety guardrails. Several achieve similar accuracy, but differ substantially in cost, latency and how often they refuse to classify a request. We explore what those differences mean when choosing a model for an application.
TypeSafe does not publish public benchmark scores for Jev, so we ran six public safety benchmarks ourselves. On accuracy, Jev sits in the same band as frontier LLMs; on cost and latency, it is nowhere near them.
Every governance check starts with the same question: what is this conversation actually saying? ClaimExtractor turns free-form AI conversations into atomic, decontextualized claims and intents you can audit, fact-check, and route.
Every AI needs a perimeter. ScopeGuard decides whether a query belongs to your AI service. Fast, multilingual, declarative, and backed by evidence for routing and analytics.