# Probabilistic Guardrails for Agentic AI
**URL:** https://discourse.julialang.org/t/probabilistic-guardrails-for-agentic-ai/139669
**Category:** General Usage
**Created:** [September 25, 2026, 11:38am UTC](https://discourse.julialang.org/t/probabilistic-guardrails-for-agentic-ai/139669 "2026-09-25T11:38:50Z")
**Posts on this page:** 2
**Page:** 1
### Author:  [@mostafatouny](https://discourse.julialang.org/u/mostafatouny)
#### Post date: [September 25, 2026, 11:38am UTC](https://discourse.julialang.org/t/probabilistic-guardrails-for-agentic-ai/139669/1 "2026-09-25T11:38:50Z")
Hello
I am writing this as an invitation to bring the toolkit of _Actuarial Science_ like _probabilistic bounds_ and _confidence intervals_ as guardrails for Agentic AI and Modern AI more broadly.
**Background**
- [Conformal Prediction](https://en.wikipedia.org/wiki/Conformal_prediction).
- [Actuarial Science](https://en.wikipedia.org/wiki/Actuarial_science) and [Value at Risk](https://en.wikipedia.org/wiki/Value_at_risk).
- [Probabilistic methods for uncertain reasoning](https://en.wikipedia.org/wiki/Artificial_intelligence#Probabilistic_methods_for_uncertain_reasoning).
**Advanced Expositions**
- [Uncertainty Quantification in LLM Agents: Foundations, Emerging Challenges, and Opportunities](https://agentuq.github.io/).
- [Advanced Topics in Statistical Learning](https://www.stat.berkeley.edu/~ryantibs/statlearn-s24/).
**Naive Example**
in an investment authority, a senior portfolio manager may set _Value at Risk (VaR)_ then a junior executes within the given scope. Similarly, an agent may be sandboxed to ensure it takes autonomous actions only if total _VaR_ is below a given threshold. In _Credit Scoring_ for example, take n to be the number of loans or transactions, and p to be the probability of default by a historical average. We can use the Percentile of a Binomial Distribution to upper bound the number of _defaults_ with a probability lower bound. If n = 1000 and p = 0.03 then at most 40 defaults may happen with confidence 0.97.
**Thesis**
AI guardrail engineering is commonly tackled by _schema checking_ and _classifiers_. Nonetheless, autonomy scalability is limited as Modern AI like Agentic AI is intrinically probabilistic. On the other hand, _Actuarial Science_ is a well-established domain for risk management under uncertainty. I am initiating the community here to bring its toolkit to design safety guardrails for modern AI. I believe [Jev](https://jevmodel.org/) is an excellent step and I can imagine beautiful use-cases if it got combined with Probabilistic Reasoning like [Probabilistic Logic](https://en.wikipedia.org/wiki/Probabilistic_logic).
**Why Julia?**
The community here knows AI as science. We have a mature native ecosystem, notably [Julia Actuary](https://juliaactuary.org/), [Turing](https://turinglang.org/), and [POMDP](http://juliapomdp.github.io/POMDPs.jl/latest/).
**Discussion**
- Do you see potential in bringing _Actuarial Science_ toolkit to modern AI safety? Are there critical challenges?
- Are you aware of any community following that direction?
- If you were to initiate a new direction, what would be your first steps?
---
### Author:  [@obsidianjulua](https://discourse.julialang.org/u/obsidianjulua)
#### Post date: [September 25, 2026, 1:30pm UTC](https://discourse.julialang.org/t/probabilistic-guardrails-for-agentic-ai/139669/2 "2026-09-25T13:30:08Z")
> [@mostafatouny](#):
>
> I believe [Jev](https://jevmodel.org/) is an excellent step and I can imagine beautiful use-cases if it got combined with Probabilistic Reasoning like [Probabilistic Logic](https://en.wikipedia.org/wiki/Probabilistic_logic).
I was reading the first part of this post thinking π€ βis he talking about jev?β because Im down the rabbit hole with this one in spectacular fashion for whatever reason and then you mentioned jev and I agree with all of this. I had to look up actuarial though and needed that link.