# Hallucination

`kaal:entity:hallucination`

**Status.** derived

This node is assembled mechanically from the 5 claims that carry the concept tag `hallucination`. It is a roster of what the corpus says under this term. It is **not** an adjudicated definition: no single statement here has been ruled canonical, and no first-appearance call has been made. Read the claims and judge for yourself.

## Every claim under this term

5 claims across 2 works, 2025 to 2026.

**2025**

- [5541658-007](https://wulfkaal.github.io/claims/5541658-007) [failure/evidenced] *(failure mode)* -- Retrieval-augmented generation reduces but does not eliminate hallucination: leading legal research tools including Lexis+AI and Westlaw AI-Assisted Research still generate false citations or facts in 17 to 33 percent of cases.
  > the risk of "hallucinations," where AI generates false legal citations or facts, as observed in tools like Lexis+AI and Westlaw AI-Assisted Research, which hallucinate in 17–33% of cases.
  Wulf A. Kaal, Morgan A. Gray, The Evolving Role of Artificial Intelligence in Law (2025). SSRN: https://ssrn.com/abstract=5541658

**2026**

- [6421319-021](https://wulfkaal.github.io/claims/6421319-021) [failure/argued] *(failure mode)* -- If probabilistic AI agents cannot reliably distinguish truth from fabrication, the lemons problem and signaling games retain relevance, but the asymmetry relocates: it now runs between agents that can verify AI output and those that cannot.
  > If probabilistic AI agents cannot reliably distinguish truth from fabrication, then Akerlof's lemons problem and Spence's signaling games retain their relevance. Only now the asymmetry is between agents that can verify AI output and those that cannot.
  Wulf A. Kaal, The Collapse of Scarcity Economics (2026). SSRN: https://ssrn.com/abstract=6421319
- [6421319-022](https://wulfkaal.github.io/claims/6421319-022) [empirical/evidenced] -- Hallucination rates vary sharply by domain: leading frontier models achieve sub one percent rates for general knowledge queries, while rates climb to five to thirty percent for specialized domains and legal information hallucination averages 6.4 percent even for top models.
  > While leading probabilistic frontier models have achieved sub-1% hallucination rates for general knowledge queries, rates climb to 5–30% for specialized domains, and legal information hallucination rates average 6.4% even for top models.
  Wulf A. Kaal, The Collapse of Scarcity Economics (2026). SSRN: https://ssrn.com/abstract=6421319
- [6421319-023](https://wulfkaal.github.io/claims/6421319-023) [mechanism/argued] *(failure mode)* -- Probabilistic language models hallucinate because they are trained to predict statistically likely token sequences rather than to verify propositional truth, so the error is intrinsic to the substrate: plausibility and truth are orthogonal properties in high dimensional token space.
  > Probabilistic large language models hallucinate because they are trained to predict statistically likely token sequences, not to verify propositional truth.
  Wulf A. Kaal, The Collapse of Scarcity Economics (2026). SSRN: https://ssrn.com/abstract=6421319
- [6421319-024](https://wulfkaal.github.io/claims/6421319-024) [design/argued] -- Hybrid neuro symbolic architectures structurally shrink the hallucination surface because the probabilistic layer generates candidate outputs while a symbolic layer filters, validates, and rejects any output violating formally specified rules, known ontological relations, or deductively verifiable constraints.
  > The probabilistic layer generates candidate outputs; the symbolic layer filters, validates, and rejects any output that violates formally specified rules, known ontological relations, or deductively verifiable constraints.
  Wulf A. Kaal, The Collapse of Scarcity Economics (2026). SSRN: https://ssrn.com/abstract=6421319

## Verify

Every claim above resolves to a record carrying a verbatim source quote, the sha256 of the source PDF, and a preformatted citation. Nothing here asks to be taken on trust.

    curl -s https://wulfkaal.github.io/entities/hallucination.md | sha256sum

**Canonical form.** This markdown file is the canonical hashed representation of this entity node. Its sha256 is the content hash.