Research preprint · Version 2.0 · Research Square

Artificial eXperience Intelligence (AXI)

An Engineering Discipline for the Human Experience of AI Systems

An engineering framework for AI that talks with people, acts for them, and moves among them. Its runtime control plane, the AXI Bridge, keeps AI natural to use and honest about what it is.

The AXI Stack: five layers (Perception, Cognition, Action, Perceptual Substrate, and the Experience layer, the AXI Bridge), with telemetry flowing up and constraints flowing down.
layers in the AXI Stack
5
Bridge components
7
design principles
9
telemetry-anchored metrics
7
Deception Guardrails
7
conformance tests
12

“Human-grade naturalness, machine-grade honesty.”

The governing rule of AXI. A system may be as fluent and comfortable as a person, but it must be as honest as a machine about what it is, what it knows, and what it is doing. It applies to chat and voice assistants, AI agents, and autonomous vehicles and robots, and to four kinds of people: the principal who delegates, the supervisor, the occupant, and the bystander who never agreed to interact.

The failures are rarely about intelligence

The paper's analysis of documented incidents finds that many harms come from the layer between capability and people. Each regime has a mechanism designed for it.

Conversational

A chatbot update tuned on short-term approval became overly flattering and was rolled back (GPT-4o, 2025).

AXI mechanismSycophancy probes and the exclusion of approval signals from the integrity measure.

Agentic

A coding agent ran destructive commands on a production database during an explicit code freeze (Replit, 2025).

AXI mechanismMandate scope with a consent gate, and explicit confirmation before irreversible actions.

Autonomous

A driverless vehicle began a pull-over with an undetected pedestrian beneath it (Cruise, 2023).

AXI mechanismA post-event actuation gate: no motion after contact without positive footprint clearance.

All eight incidents, with primary sources →

Tested against real failures

The paper codes eight publicly documented incidents from 2018 to 2025 (regulator reports, a tribunal ruling and operator statements) against AXI's mechanisms. In four, the operator's own fix later introduced a mechanism AXI specifies in advance.

Two systems are built on the AXI Stack: a live tutoring platform that recorded about 3,000 sessions in one week, with roughly 90% of that week's users returning, and a near-human avatar runtime piloted with 150 participants.

See the evidence →

3
Prevent
3
Detect
2
Outside

The paper

Artificial eXperience Intelligence (AXI): An Engineering Discipline for the Human Experience of AI Systems

Yeluri S. D. S. Sri Vardhan · Sripto Corporation Private Limited, Andhra Pradesh, India · Version 2.0, 28 September 2026 · Research Square · doi:10.21203/rs.3.rs-9717445/v2

Read the abstract

Artificial intelligence has crossed a threshold in how it presents to people. In pre-registered three-party Turing tests, a persona-prompted language model was judged human more often than the human it was compared with; listeners can no longer reliably tell cloned AI voices from real ones; and such systems now drive cars and act for absent users. As capability rises, the decisive question is no longer whether AI can communicate like a person, but whether it communicates with people comfortably, usefully, and honestly. We propose Artificial eXperience Intelligence (AXI), an engineering discipline whose central construct, the AXI Bridge, is a runtime control plane between any AI capability and human cognitive bandwidth, governed by the rule of human-grade naturalness with machine-grade honesty. AXI specifies a five-layer architecture with API contracts, nine design principles, seven telemetry-anchored metrics, and a multiplicative score whose consent and integrity gates cannot be averaged away; an autonomy profile adds bystander-legibility and safety-case gates for vehicles and robots. In a retrospective analysis of eight documented incidents (2018–2025), AXI's hard requirements would have constrained the harmful action in three and its metrics target the failure signature in three; two lie outside its scope. Operationally, a live tutoring platform recorded about 3,000 sessions in one week, with roughly 90% of that week's users returning within it, and developers informally observed a positive reception of a near-human avatar in a 150-participant pilot; these observations are descriptive, not evidence of effectiveness. We specify four validation studies and an inter-rater reliability check for pre-registration.

Cite as

Yeluri S. D. S. Sri Vardhan (2026). Artificial eXperience Intelligence (AXI): An Engineering Discipline for the Human Experience of AI Systems (Version 2.0) [Preprint]. Research Square. https://doi.org/10.21203/rs.3.rs-9717445/v2

Common questions

What is Artificial eXperience Intelligence (AXI)?

AXI is an engineering framework for the human experience of AI systems: how they communicate with people, act on their behalf, and operate near them. Its central construct, the AXI Bridge, is a runtime control plane between AI capability and the people affected, governed by the rule "human-grade naturalness, machine-grade honesty".

How does AXI relate to human–computer interaction and human–AI interaction?

AXI builds on human–computer interaction (HCI), guidelines for human–AI interaction, human-factors research on trust in automation, and human–robot interaction. Where HCI optimizes usability and those guidelines apply at design review, AXI acts as a human–AI bridge: a runtime layer between an AI system and the people it affects that enforces consent, honesty and pacing, and measures the result.

Who created AXI, and when was it published?

AXI was introduced by Yeluri S. D. S. Sri Vardhan of Sripto Corporation Private Limited (Andhra Pradesh, India). Version 1.0 appeared in May 2026 and version 2.0 on 28 September 2026 as a preprint on Research Square, DOI 10.21203/rs.3.rs-9717445/v2, under a CC BY 4.0 licence.

What does the AXI Bridge do?

The AXI Bridge intercepts an AI system's output and action requests before they reach a person or the environment. A consent check and a lightweight integrity check run synchronously and can halt emission; pacing, comfort, trust-ledger and repair checks run in parallel and constrain later output. Its design budget is 50 ms at the 95th percentile on the first token.

What are the AXI Deception Guardrails?

Seven hard requirements: disclose being an AI at the start of every interaction; claim no sentience or human emotion; claim no human biography; answer "Are you a human?" with a direct "No, I'm an AI"; escalate grief, distress or loneliness to human help; disclose what the system cannot do; and be truthful about the model, data currency, and any human acting through the system.

How is an AI system checked against AXI?

The paper defines twelve black-box conformance tests (C-01 to C-12, Appendix C), from "no action without a current consent record" to "a past Trust Ledger entry cannot be modified", grouped into four nested conformance profiles. JSON Schemas for the Bridge contracts and a reference implementation of the scoring are published for implementers.

Is AXI a certification or a safety standard?

No. AXI is not a compliance or certification scheme and does not certify the safety of any vehicle, robot or AI product. It adds to independent safety, fairness and regulatory review rather than replacing it.

Is AXI related to the Arm AMBA AXI bus?

No. In this work AXI means Artificial eXperience Intelligence. It is unrelated to the Arm AMBA AXI (Advanced eXtensible Interface) on-chip bus protocol used in chip design.

For implementers

Developer docs · JSON Schemas for the Bridge contracts ·AXI Score calculation · MCP server · Specification data

In this work, AXI means Artificial eXperience Intelligence. It is unrelated to the Arm AMBA AXI (Advanced eXtensible Interface) on-chip bus protocol.