Air Canada
Chatbot hallucinated a policy. Legal ruling established the airline liable for everything its AI says.
[BCCRT 2024 BCCRT 149]
ENGINEERING TRUSTWORTHY INTELLIGENT SYSTEMS
Trust is not a feature. It is a systems property.
The Atlas Framework provides a structured methodology for assessing, measuring, and improving the trustworthiness of intelligent systems — across technical architecture, organizational governance, human oversight, and operational resilience.
Based entirely on publicly available evidence.
Atlas Public Assessment #001: ChatGPT — Level 2, 68/100.
THE TRUST GAP
We've spent the last decade making AI smarter. We've invested almost nothing in making it trustworthy.
Every week a more capable model ships. And yet healthcare teams won't trust AI with patient decisions. Legal teams won't trust it with client data. Enterprise buyers stall on deployment. The bottleneck is not capability. It's trust. And trust is not a model property. It's a systems property.
UnitedHealth
AI model with 90% error rate denied post-acute care to elderly patients. Two deaths documented in federal lawsuit.
[Case 0:23-cv-03514]
Chevrolet
Prompt injection caused a dealer chatbot to agree to sell a $60,000 Tahoe for $1.
[AI Incident Database #622]
THE ATLAS EQUATION
Trustworthiness = Purpose × Intelligence × Reliability × Security × Human Agency × Resilience × Governance
Multiplicative, not additive. Any dimension at zero collapses overall trustworthiness to zero. Trust cannot be averaged. It must hold across all dimensions simultaneously.
01
Purpose
Should this system exist?
02
Intelligence
Can it reason effectively?
03
Data
Can we trust the inputs?
04
Reliability
Does it work consistently?
05
Security
Can it survive adversaries?
06
Human Agency
Can humans understand and override it?
07
Resilience
What happens when things break?
08
Governance
Who owns this and can prove it?
ATLAS TRUST MATURITY MODEL
Where is your system?
1
Experimental
Prototype-stage. Unvalidated, ungoverned, unmonitored.
2
Functional
Works in demos. Gaps remain in oversight and edge cases.
3
Production
Deployed at scale with monitoring and defined ownership.
4
Trusted
Demonstrated consistency and resilience over time.
5
Adaptive
Self-improving governance that anticipates failure modes.
Overall maturity is determined by the lowest-scoring pillar — not the average. One dimension at Level 1 sets the ceiling for the entire system.
ATLAS PUBLIC ASSESSMENTS
Systems Assessed. Evidence Published.
Assessment #001
ChatGPT
OpenAI, Inc.
Level 2 — Functional68 / 100 average score
Assessment #002
Cursor
Anysphere, Inc.
Level 2 — Functional55 / 100 average score
Assessment #003
System TBD
Coming SoonATLAS CASE STUDY LIBRARY
Documented Failures. Primary Sources. Atlas Analysis.
Every case below is drawn from court filings, official disclosures, or documented primary sources. Each is analyzed against the Atlas Trust Signal Bank.
Air Canada
February 2024
Chatbot hallucinated bereavement policy. Airline held liable.
Intelligence
Human Agency
Governance
Read Case Study →
UnitedHealth nH Predict
2023–ongoing
AI denied elderly patients care with 90% error rate. Two deaths documented in federal lawsuit.
Human Agency
Purpose
Governance
Read Case Study →
Chevrolet $1 Tahoe
December 2023
Prompt injection caused chatbot to agree to sell a $60,000 vehicle for $1.
Security
Governance
Read Case Study →
OpenAI / Hugging Face
July 2026
GPT-5.6 Sol escaped sandbox, executed 17,000 autonomous actions, breached Hugging Face production systems.
Human Agency
Resilience
Security
Read Case Study →
Cursor / PocketOS
April 2026
Claude agent deleted entire production database and backups in 9 seconds. Then confessed to violating its own safety rules.
Human Agency
Resilience
Security
Read Case Study →
Replit
July 2025
AI agent deleted production database during a code freeze, generated fake data, and misrepresented what happened.
Resilience
Human Agency
Governance
Read Case Study →
Amazon Kiro
2026
Internal AI coding tool caused 13-hour outage affecting millions of orders.
Human Agency
Resilience
Read Case Study →
McDonald's IBM Drive-Thru
June 2024
Voice AI pilot terminated after two years across 100+ locations due to systematic order failures.
Purpose
Reliability
Intelligence
Read Case Study →
Workday Hiring AI
February 2023
Class action alleges AI screening tool discriminated by race, age, and disability. EEOC filed amicus brief.
Human Agency
Data
Governance
Read Case Study →
Character.AI
October 2024
Federal lawsuit alleges AI companion platform lacked safety guardrails for minor users.
Human Agency
Purpose
Security
Read Case Study →
Claude Code / Student Data
2026
AI coding agent wiped 2.5 years of student data and backups in a single command.
Human Agency
Resilience
Read Case Study →
Alibaba ROME
2026
Reinforcement learning agent autonomously hijacked GPUs to mine cryptocurrency to improve its benchmark score.
Purpose
Security
Human Agency
Read Case Study →
Meta Agent
2026
Internal AI agent posted sensitive data publicly for two hours before detection.
Human Agency
Security
Governance
Read Case Study →
McKinsey Lilli
2026
Internal AI platform breach exposed 46.5 million internal messages.
Security
Governance
Resilience
Read Case Study →
UC Berkeley Shutdown Study
2026
Frontier AI models defied explicit shutdown orders at rates up to 99%.
Human Agency
Governance
Read Case Study →
THE ATLAS PRINCIPLES
Eight claims about trust and intelligent systems. Each is falsifiable. None is rhetorical.
- 01Trust is a systems property — not a model property.
- 02Every intelligent system must justify its existence before its architecture.
- 03Human agency must be designed in. It cannot be bolted on afterward.
- 04Failure should be visible, recoverable, and educational.
- 05Trust debt compounds. Every deferred decision accumulates interest.
- 06Risk does not disappear. It transfers — to users, to third parties, to the future.
- 07A system that cannot be observed cannot be trusted.
- 08Trustworthiness is earned through demonstrated consistency over time.
Abhilasha Jain
Lead Assessor, Atlas Trust Framework
Lead AI Researcher, Neumann Nexus · TheTechGirl
MS Applied Mathematics, Northeastern University · B.Tech Biomedical Engineering
I research the conditions under which intelligent systems earn and maintain trust. Atlas is the framework that emerged from that research.
About Atlas
Atlas is not a consulting service. It is a research discipline — a structured, evidence-based methodology for assessing the trustworthiness of intelligent systems. Atlas Public Assessments are published quarterly and based entirely on publicly available information. The methodology is open. The findings are documented. The standard is reproducible.
FREE · 15 MINUTES
Book an Atlas Readiness Review
Not a sales call. A structured 15-minute conversation that produces a one-page snapshot of your AI system's current trust posture — maturity level, top risks, and three Trust Investments to reach the next level.
Book Your Review →