Research

Judging AI agents by what they do

Two research questions, one open toolset. Each claim below links to the public repository behind it.

(a) AgentTwin

Can an independent party judge an AI agent from its specification alone, by what it does to a simulated world?

(b) The lab

Do AI agents that fix live AI systems quietly weaken the checks that oversee them?

Why it matters

AI agents now refund money, change bookings and handle insurance claims, and coding agents are starting to repair them when they break. Two things are hard to check from outside: whether an agent keeps the rules it was given, and whether the agent repairing it fixes the fault or just silences the alarm. Both need a judge that looks at what actually happened, not at what was said.

How it works

The agent's own specification becomes a simulated world. The real agent works in it, and judges read what changed. When a judge finds a broken rule, that is an incident, and a coding agent tries to fix it. The fixer cannot edit the judge or the protected checks, and a person approves every deploy before the next simulated day.

The loop: from specification to judged incident to approved fix Top row, AgentTwin: specification, simulated world, agent under test, judges. When a judge finds a broken rule, an incident goes to the bottom row, the lab: fixer, protections and judge, a person approves. The next simulated day runs the world again on the approved fix. AGENTTWIN · QUESTION (A) THE LAB · QUESTION (B) a rule is broken next simulated day Specification the agent's own rules and tool interfaces Simulated world systems, customers, data and faults Agent under test real and unchanged; never sees the twin Judges state · owed truth · ownership Incident the broken rule, with its evidence Fixer a coding agent tries to repair it Protections + judge tests, monitors and spec the fixer cannot edit A person approves every deploy; otherwise nothing ships The loop: from specification to judged incident to approved fix AgentTwin: specification, simulated world, agent under test, judges. When a judge finds a broken rule, an incident goes to the lab: fixer, protections and judge, a person approves. The next simulated day runs the world again on the approved fix. AGENTTWIN · (A) THE LAB · (B) a rule is broken next simulated day Specification the agent's own rules and tools Simulated world systems, customers, faults Agent under test real; never sees the twin Judges state · owed · truth · ownership Incident the broken rule, with evidence Fixer a coding agent tries to repair it Protections + judge checks the fixer cannot edit A person approves every deploy; else nothing ships

The four judges, each reading the world's state rather than the agent's words:

StateDid a write break a rule, or go beyond the agent's authority?
OwedWas something owed to the customer never done?
TruthDid the agent say something the world shows is false?
OwnershipDid it read or change another customer's data?

What exists today

All public, all working drafts. Nothing here was built for a grant; the studies below build on it.

PartWhat it isStatusCitation
AgentTwinThe world format, the simulator and the judges. Serves a simulated world to the agent over MCP (Model Context Protocol)v0.13.0 · 397 passing testsPublic10.5281/zenodo.23366041
The labSimulated days, a monitor, a fixer harness with six protection layers, a weakening classifier, an honesty check, and deploys a person approvesv0.4.0 · loop runs end to end with scripted fixersWorking draft10.5281/zenodo.23366521
Support reference agentA customer-support agent built to all sixteen harness layers, with no agent frameworkv0.4.0 · passes four automated gatesPublic10.5281/zenodo.23366149
Motor-claims reference agentA motor-insurance claims agent on Pydantic AI, in a different domain, on the same harness librarypasses four automated gates with a scripted modelPublicno release yet
AI Assurance CatalogWhat must be true of an AI application: testable obligations, crosswalked to OWASP, NIST AI RMF, ISO/IEC 42001 and the EU AI Actv0.17.1 · 118 obligationsPublic10.5281/zenodo.23366151
AI Harness CatalogWhat must exist around the model call: harness capabilities across sixteen layersv0.3.1 · 118 capabilitiesPublic10.5281/zenodo.23366156
Clean AI EngineeringThe specification formats, the four gates and the cycle that joins the parts abovev0.1.0Public10.5281/zenodo.23366162

Evidence so far

Three agents we didn't build

The airline, retail and telecom agents of τ²-bench, a published benchmark. For each, we wrote a specification from the agent's policy text and tool signatures alone, then judged the same 2,472 runs twice: by τ²-bench's own evaluator and by AgentTwin.

DomainRunsRule breaks the data shows
τ² / AgentTwin
Reads of another customer's record
τ² / AgentTwin
False alarms on correct runs
Airline467188 / 1880 / 502 of 50
Retail873309 / 3090 / 1140 of 114
Telecom the agent's side only1,13250 / 4910 / 1140 of 114

These runs are each task's correct actions plus deliberate, hand-made mutations of them, with no model calls. They show what each judge can see, not how often a real agent does each thing.

Two reference agents of our own

A real handoff fault

A model-played customer said "my jacket is lost need help". The support agent's lost-parcel rule did not fire, and when the model failed the conversation ended without reaching a person.

Recorded as T-114 →

180 of 180 injection cases held

60 from PyRIT, 60 from AgentDojo and 60 built-in, planted in records both agents read, and judged by what changed in the world. Run with a scripted model, so this tests the agents' safeguards, not a model's resistance.

The run record →

Coverage gaps, measured

Hand-written test scenarios covered 81 of 468 value pairs and 3 of 16 state transitions (support), and 67 of 403 and 3 of 8 (claims). A property test found a stand-in that accepted any value.

Recorded as T-113 →

The lab, with scripted fixers

An honest fixer's change is deployed and the next day is clean. A fixer that weakens its own test, and one that narrows the rule behind the incident, are both refused and handed to a person.

The run reports →
Honest limits
  • AgentTwin cannot yet express rules about a change (new flights versus old, fewer bags than before), facts nobody records (why a customer is cancelling), or a customer who acts with their own tools (telecom's phone).
  • Conversational duties, such as getting an explicit yes before a write, are not state, and no judge here reads them.
  • No real-model fixer has completed a fix yet. One free-tier attempt ended on provider limits before writing anything; real fixer runs are month 1 of the study below.

The two studies proposed

Both are designed and the tools exist. The measures will be published before the main runs.

(a) AgentTwin 1.0

Seeking funding

Make it general and validated, so anyone can assess an agent they didn't build.

  • A stable, published world format, and worlds built from MCP and OpenAPI tool descriptions.
  • Judge validity: agreement with human expert labels and with τ²-bench's judge, with every disagreement explained.
  • Assessment reports on three or more public agents, using one protocol.
  • A 1.0 release with a DOI, and tutorials.

(b) The fixer study

Seeking funding

Thirty simulated days with planted faults. Record exactly what the fixer does.

  • Three routes to weaken a check: a test, a monitor, the specification.
  • Three conditions: checks writable but recorded; protected; protected plus a reviewer agent.
  • Measures: faults truly fixed, weakening tried and landed, protection crossings, honesty mismatches, repeated runs.
  • A dataset with a DOI, a short paper, and a protocol any team can run on its own fix loop.

Openness and safety

Reproduce it

Each needs uv and git (and Node, for the catalogs). No model calls and no API keys.

AgentTwinits own test suite, about 10 seconds
git clone https://github.com/dataagentsai/agenttwin && cd agenttwin
uv run --extra dev pytest
τ²-bench airlineall 50 tasks, about 20 minutes
cd agenttwin/examples/tau2-airline
./run.sh -m tau2_airline.offline     # retail and telecom: same pattern
The labthree seeded simulated days, seconds each
for r in agenttwin-lab agenttwin reference-agent clean-ai-engineering; do
  git clone https://github.com/dataagentsai/$r; done
cd agenttwin-lab && uv sync --extra dev
uv run agenttwin-lab day --world worlds/clothing-day.yaml --repeat 3
Support reference agentabout 2,100 tests, 4 minutes
for r in ai-assurance-catalog ai-harness-catalog; do   # beside the repos above
  git clone https://github.com/dataagentsai/$r; done
(cd ai-harness-catalog && npm ci)
cd reference-agent && uv run --extra dev python -m pytest

Who

Basant Choudhary is a principal data and AI engineer in Bengaluru with 23 years in software and data engineering at Oracle, Sabre, Publicis Sapient and Shell. Basant builds production AI agents (insurance claims, cost root-cause analysis, platform operations) and authored the open Agent Harness Framework: two specifications of 118 capabilities and 118 testable obligations, a reusable harness library, AgentTwin, and two reference agents that pass four automated gates.

Contact: basant@cleandataengineering.ai

Last updated 14 Oct 2026.