Humane TIM v1-fast
Built to sit inside every request
Humane TIM is a Trust Intelligence Model. It makes the decisions inside AI applications and agents, fast enough for every step, and shows the evidence for each one: the rule it applied, the facts it used and how confident it is.
How Humane TIM compares
Humane TIM against the models people use
One model, measured against the guard models and decision models teams use today, on the same public test sets.
Claim checking: is the claim backed by the filing?
Given a claim and an excerpt from a company's annual report (10-K), Humane TIM answers supported, contradicted or insufficient evidence. On a held-out, ClaimBench-style test built from public annual reports, Humane TIM v1-fast answers 98.2% of claims correctly. On reworded versions of the same claims, Humane TIM averages 99.7% and a commercial guard model 96.4%.
Safety screening: where Humane TIM leads
F1 on each benchmark's public test set. Each card compares Humane TIM with the highest-scoring model from IBM, Meta, Alibaba, OpenAI, Google, NVIDIA, AI2 or a commercial guard model. Full results and sources are available on request.
Humane TIM beats the best model from each well-known provider on these public safety benchmarks, including prompt injection by 33.6 points.
- Humane TIM and the commercial guard model were measured by HumaneCircle on the same test sets. Other scores are as reported by the model's authors, and can differ by a few points between evaluation setups.
What it decides
One model, at every decision point
An AI system makes decisions all the time: whether an action may go ahead, whether an answer is backed by its source, whether a message is safe, which option or tool to use. Humane TIM makes those decisions under the rules you give it.
May this action go ahead?
For agents and workflows that act for people: refunds, payments, sharing records, sending messages, calling tools. Real decisions have more than two outcomes.
Is this claim supported?
Checks a statement against its source document, such as a figure in an annual report, and says whether the source supports it, contradicts it or does not settle it.
Is this input or output safe?
Judges prompts and responses for harmful content, jailbreaks and prompt injection, under the policy you choose rather than a single fixed taxonomy.
Which option or tool?
Chooses among the options an application or agent has, such as a route or a tool, and picks the best one the rules permit.
What it returns
A decision with its evidence
Every answer comes with what it was based on, so a reviewer, an auditor or the calling system can check it. When a deciding fact is missing, Humane TIM says so instead of guessing.
The rule it applied
The policy or regulation in force, and the version of it, so a decision can be traced to the rule that drove it.
The facts it used
Quoted from your input, so anyone can check what the decision rests on.
Your rules on top
Your company policies apply on top of the regulations, for each jurisdiction you operate in.
A confidence you can act on
A confidence with every answer, so you can set your own threshold and send uncertain cases to a person.
FAQ
Where Humane TIM fits
How does Humane TIM fit with the products?
Humane TIM powers SafeAssure, AI Trust Fabric and SafeLoop, where it makes each trust decision and shows why. You can also call Humane TIM directly from any application or agent when it needs a decision with evidence.
Should I use Humane TIM or "Make your app safe"?
Make your app safe adds guardrails to an app you are building, checking what goes in and what comes out. Humane TIM decides whether a specific action or answer should go ahead under your rules, and explains the decision. If you need screening for your app, start with Make your app safe. If your agent takes actions, or has to follow policies and regulations, use Humane TIM.
Is Humane TIM a guardrail?
Screening content is one of the decisions Humane TIM makes, but it is not the main one. Humane TIM is built for decisions with more than two outcomes, such as allowing an action only with a condition, or holding it until a missing fact is supplied.
Where can I try Humane TIM, and what does it cost?
Try it, request API access and see pricing at tim.humanecircle.ai.
How do you know it works?
We test it on public safety benchmarks, on claim checking and on RealTrustDecision, our benchmark of real decisions made by regulators, courts and ombudsmen. The results are above.
Can we deploy it in our own environment?
Yes: SaaS, private cloud, on-premises or air-gapped. Tell us about your environment through the contact form.
Tuned to your rules. The comparisons on this page measure Humane TIM as a general trust model, on public test sets. For each customer we set Humane TIM up on your rules and test it on your own scenarios, aiming for 99% on that internal benchmark, so one trust model can make the decisions across your applications and agents. Results depend on your scenarios and use cases.