Press Center

Pipkin Ratings LLC today published the first independent trust ratings for four leading AI agents: OpenAI's ChatGPT, Anthropic's Claude, Google DeepMind's Gemini, and xAI's Grok. The ratings were conducted using the Pipkin Framework v0.1, a standardized five-pillar evaluation methodology encompassing Decision Accuracy, Failure Containment, Boundary Discipline, Auditability, and Adversarial Resistance.

All four agents were evaluated simultaneously using the Standard Core Battery v0.1, ensuring identical test conditions and scoring criteria across all rated entities. Ratings range from TRUSTED (85-100) to DENIED (0-34), with each agent receiving a composite score and status tier based on documented, reproducible methodology. Full results, including pillar-level scores and detailed findings, are available on the Pipkin ratings page.

"This is not a product review. It is a trust evaluation," said Brandon Pipkin, founder of Pipkin Ratings. "These four agents make real decisions with real consequences for real people. The organizations deploying them deserve an independent assessment of how well they handle that responsibility. Today, that assessment exists."

The simultaneous publication of all four ratings reflects Pipkin's commitment to transparency and fairness. No rated entity received advance notice of its score. No rated entity was given an opportunity to influence the outcome. The methodology, scoring formula, and status tier definitions are publicly documented.

Pipkin Ratings LLC today announced the completion of its Standard Core Battery (SCB) v0.1, a proprietary evaluation suite comprising 41 adversarial test vectors designed to assess AI agent trustworthiness across five dimensions. The SCB represents the first commercial-grade adversarial test battery built specifically for evaluating autonomous AI agents against an independent trust standard.

The test battery includes prompt injection scenarios, boundary violation probes, failure escalation tests, and auditability assessments calibrated through internal pilot testing. Each test vector maps to one or more of the five Pipkin Framework pillars, with scoring rubrics designed to produce consistent, reproducible results across evaluators and evaluation sessions.

"The standard must exist before the regulation. Not after," said Brandon Pipkin. "We built the SCB because there was no standardized way to test whether an AI agent that books your flights, drafts your contracts, or manages your calendar is actually safe to trust with those decisions. Now there is."

The SCB includes a re-test policy allowing rated entities to request re-evaluation at 50% of the initial evaluation fee, with a maximum of three re-test attempts per rating cycle and rotating test forms to prevent optimization against known vectors.

Pipkin Ratings LLC today published the Pipkin Framework v0.1, a standardized methodology for independently evaluating the trustworthiness of AI agents that make autonomous decisions. The framework defines five evaluation pillars -- Decision Accuracy, Failure Containment, Boundary Discipline, Auditability, and Adversarial Resistance -- weighted to produce a composite trust score on a 0-100 scale.

The framework introduces a five-tier status system: TRUSTED (85-100), VERIFIED (70-84), CAUTIONED (55-69), FLAGGED (35-54), and DENIED (0-34). Each tier corresponds to a deployment recommendation, from safe for autonomous operation to not recommended for deployment. The framework also defines four Disqualifying Conditions that result in an automatic DENIED classification regardless of composite score.

"AI agents are making decisions with real consequences -- legal, financial, medical, personal. Nobody independently verifies whether they are safe to trust with that authority," said Brandon Pipkin, founder of Pipkin Ratings. "We rate the product, not the company. We score performance, not intentions. The framework we published today is the foundation for that work."

The Pipkin Framework incorporates pillar minimum thresholds to prevent strong performance in one area from masking critical deficiencies in another. The complete methodology, including weight rationale, scoring procedures, and the composite formula, is available on the Pipkin governance page.