VERITAS
The audit log

Notes from the verification layer.

Short pieces on pre-filing citation verification, Verification Certificates, and the discipline a firm adds when its drafting workflow now includes generative tools.

2026-07-13

What the Certificate Looks Like in a Sanctions Hearing: A Counterfactual on Mata v. Avianca

Going back to Mata v. Avianca and reconstructing what the order to show cause looks like if a Verification Certificate had been filed with the brief. The certificate is not protection. It is the record.

2026-07-06

Malpractice Underwriting in the Generative AI Era: What Carriers Are Asking on Renewal

What major LPL carriers have started asking about AI use, citation verification policies, and supervision protocols. The Verification Certificate is the artifact that converts representation into evidence.

2026-07-01

Deterministic vs. Probabilistic Citation Checks, Explained Simply

A plain-English explainer on the two ways a computer can check a legal citation. A probabilistic check guesses whether a case is real. A deterministic check looks it up in the record. Where each one fails, and why a real record lookup, not a guess, should decide whether a case exists.

2026-06-29

Proposition Support: The Verification Step That Comes After Citation Existence

The harder failure mode. The case exists. The cite is real. But the case does not stand for the proposition the brief claims. Where citation verification ends and adversarial review begins.

2026-06-28

Landberg v City of New York: The Second Department's Sanction for AI-Hallucinated Citations

New York's Appellate Division, Second Department directed an attorney to pay $8,000 and his firm $2,500 to the Lawyers' Fund for Client Protection after a brief the court described as prepared with generative AI cited nonexistent cases and misstated real ones. The second appellate AI-citation sanction in three weeks, and the mechanical verification step that resolves the existence question before filing.

2026-06-22

Anatomy of a Hallucination: The Morgan & Morgan Filing and Why Internal Review Did Not Catch It

The first BigLaw-scale AI hallucination incident. Different audience: a national plaintiffs' firm with internal review processes. The story is that the firm's review did not catch the cite.

2026-06-15

Rule 11 and Generative AI: What Sanctioning Judges Have Actually Said About the Duty of Inquiry

A close reading of the sanctions opinions focused on the Rule 11(b)(2) language each court relied on. Courts are not punishing AI use. They are punishing the absence of a verification step.

2026-06-10

The Best Model Anthropic Has Shipped Still Hallucinates. Here Is What the System Card Says.

Anthropic released Claude Fable 5 with the best knowledge-and-hallucination benchmark on record, a 40 out of 100. The 319-page system card still documents residual error, and in some tests the new models hallucinate more than the older Opus. For a filed brief, the residual is the whole problem. Why a better model does not lower the verification duty, and the discipline layer that makes AI-assisted drafting defensible.

2026-06-08

What a Hedged Verdict Actually Means: The Verification Status Taxonomy

The verdict states a scanner can return, and why hedged verdicts matter more than binary ones. A confident wrong answer is the failure mode that sanctioned the Mata lawyers.

2026-06-01

Anatomy of a Hallucination: Park v. Kim and the Second Circuit Referral

The second sanctions case in the canonical sequence. A pro se attorney, lower stakes, the same mechanism as Mata. Line-by-line on the failure mode the audit layer was built for.

2026-06-01

Florida Rule 2.515: The New AI Citation Certification Effective June 15

Effective June 15, 2026, every signer of a Florida court filing represents that the legal authorities cited exist and are accurately cited. One statewide standard replaces the patchwork of trial-court AI orders. What the representation asks, what is exposed when it fails, and the pre-filing verification step that satisfies it.

2026-05-25

Federal Judge AI Standing Orders: A Working Inventory and What They Actually Require

What published AI standing orders actually demand, grouped by requirement: disclosure, certification of verification, prohibition. The operational answer to the certification requirement.

2026-05-20

Anatomy of a Hallucination: Mata v. Avianca and the Citations That Did Not Exist

Mata v. Avianca is the canonical example of an AI-hallucinated citation reaching a federal docket. A line-by-line look at what the brief cited, what could not be located in the reporter, and what a pre-filing verification step would have surfaced.

2026-05-19

The Pre-Filing Verification Step That Catches What Westlaw Does Not

Research platforms assume the cited case exists. A pre-filing citation check confirms it before the brief leaves the firm. Here is what the audit layer does that research platforms do not.

2026-05-19

What a Verification Certificate Contains (And Why It Belongs in the Matter File)

The anatomy of a Verification Certificate: hashed content, public token URL, hedged verdicts, and the matter-file artifact that malpractice underwriters look for.