EvSafety — safety-oriented research and evaluation

The question

EvSafety studies how AI systems behave as tool use and autonomy increase: what can go wrong, how it can be measured, and what a safeguard can and cannot claim. It is the newest branch and the least complete.

Frames concrete failure modes for tool-using and autonomous agents
Develops evaluation methods and documented test cases
Connects findings back to the harness work in EvAgent

What has been run

Early research. No public artefacts, evaluation reports or reproducible results have been released for this branch. Nothing on this page should be read as a validated outcome.

Evidence

Captures from the current build and runs. Slots marked as pending are not published as proof — they are waiting on a sanitised capture.

Schematic of the evaluation method: a task definition feeds an agent that uses tools, all inside a dashed scope boundary. Failure modes branch into a detection stage, which feeds an evaluation matrix, which returns feedback to the agent. The matrix cells carry no values. — How failures would be identified and scored. Schematic of the intended structure only — no method has been published and no dataset, model or task is named.This describes an intended evaluation approach. It is not a report of work performed, and it names no dataset, model, task or measurement.

Source and documentation

Public artefacts for this branch, checked before publication.

DocumentationPublic docs — no source code

Where this connects

Branches share infrastructure and people; the relationship is association, not a dependency chain.

EvAgent
Available

EvMind
Experimental

EvMMFPS
Experimental

EvMem
In development

EvStudy
In development

EvAgency
Active