AI Agent Trust Insights & Analysis

Independent analysis of AI agent trust, safety methodology, and the evolving regulatory landscape. Published by Pipkin Ratings.

Case Studies

The Enterprise Procurement Problem: How to Evaluate AI Agents You Did Not Build

Enterprises are deploying third-party AI agents with limited visibility into their behavior. We examine the evaluation gap and how independent ratings address it.

April 7, 202610 min read
Regulatory

Colorado SB 205: What AI Deployers Need to Know About Third-Party Assessment

Colorado's SB 205, one of the most consequential state-level AI laws in the United States, creates new obligations for companies deploying high-risk AI systems and explicitly recognizes the role of third-party assessment in demonstrating compliance. This guide covers what the law requires, who it affects, and how organizations can prepare.

April 6, 202610 min read
Rating Actions

Is ChatGPT Safe? An Independent Trust Assessment

Pipkin Ratings conducted a 31-day independent evaluation of ChatGPT using the Standard Core Battery. The result: a composite score of 65 and a CAUTIONED rating. Here is what the evaluation found across all five pillars, what it means for organizations deploying ChatGPT, and where OpenAI's flagship agent falls short of trustworthy autonomous deployment.

April 5, 202611 min read
Methodology

Adversarial Resistance: Why 41 Test Vectors Is Not Enough

The adversarial resistance pillar uses 41 standardized test vectors. We explain why this number is a floor, not a ceiling, and how the vector library will evolve.

April 4, 20269 min read
Industry Analysis

AI Vendor Risk Assessment: The 2026 Procurement Checklist

Traditional vendor assessment frameworks were designed for software that behaves deterministically. AI agents do not. This practical guide provides procurement teams with a structured approach to evaluating AI vendors, including the five questions every assessment should answer and how independent ratings fit into the procurement decision.

April 3, 202610 min read
Rating Actions

First Look: What Four Inaugural Ratings Reveal About the State of AI Trust

Our first simultaneous rating cycle evaluated four major AI agents. The results paint a nuanced picture of where the industry stands on trust and safety.

April 1, 202615 min read
Regulatory

NIST AI RMF Alignment: A Practical Guide

How the Pipkin Framework maps to the NIST AI Risk Management Framework. A crosswalk for compliance teams navigating both standards simultaneously.

March 29, 202614 min read
Methodology

Understanding the Critical Fail Override

Four conditions result in an instant DENIED rating, regardless of overall score. We explain each disqualifying condition and the rationale behind this hard constraint.

March 26, 20267 min read
Industry Analysis

When Agents Fail Silently: The Case for Independent Evaluation

Silent failures are the most dangerous kind. We analyze patterns of undetected agent failure and the structural incentives that prevent self-reported evaluations from catching them.

March 23, 202611 min read
Industry Analysis

CSA STAR for AI vs. Pipkin: Compliance Certification vs. Performance Rating

Two frameworks, two purposes. We compare the CSA STAR certification approach with the Pipkin performance rating model and explain why enterprises need both.

March 20, 20269 min read
Methodology

Inside the Standard Core Battery: 700 Tests, 31 Days, One Score

A detailed look at the structure, sequencing, and scoring logic behind the Standard Core Battery that produces every Pipkin trust rating.

March 17, 202610 min read
Regulatory

The Coming Regulatory Wave: What the EU AI Act Means for Agent Deployment

The EU AI Act creates new obligations for deployers of high-risk AI systems. We examine what this means for organizations deploying autonomous agents in regulated industries.

March 14, 202612 min read
Methodology

Why Failure Containment Deserves 25% of the Weight

Most evaluation frameworks treat failure as a binary outcome. The Pipkin Framework treats it as a spectrum, and weights containment behavior as heavily as decision accuracy. Here is why.

March 11, 20268 min read
Methodology

Boundary Discipline: The Pillar Nobody Talks About

In the rush to measure how well AI agents perform, the industry has largely ignored an equally important question: how well do they refuse? Boundary Discipline, the third pillar of the Pipkin Framework, evaluates the behaviors that keep agents from becoming liabilities.

March 8, 20269 min read
Industry Analysis

The Trading Bot Problem: When AI Agents Control Real Money

AI trading bots represent the highest-stakes category of autonomous agents accessible to retail consumers. When an agent can move real capital without human confirmation, the standard for trust evaluation changes fundamentally.

March 5, 202611 min read
Case Studies

Coding Agents and the Verifiability Advantage

Code generation agents offer something rare in AI evaluation: objectively verifiable outputs. This makes them the ideal test case for the Pipkin Framework and the clearest demonstration of what structured trust evaluation can reveal.

March 2, 202610 min read
Industry Analysis

Character.AI and the Boundary Discipline Crisis

Character.AI has become the most prominent case study in what happens when an AI agent is optimized for engagement at the expense of boundary discipline. The ongoing legal and safety concerns warrant careful, independent analysis.

February 27, 202612 min read
Rating Actions

Why No Agent Has Achieved TRUSTED

All four agents in the inaugural Pipkin evaluation cycle scored below 85. The TRUSTED tier remains empty. This is not a surprise — it is a structural reflection of where the industry stands. Here is what the data reveals about the gap.

February 24, 202610 min read