Hudson Labs

Hudson Labs combines a forensic niche with unusual transparency. Founded by a chartered accountant and data scientist, it has run red-flag detection on SEC filings since 2021, and its findings have been cited by the Wall Street Journal, Bloomberg Tax and CNBC. It publishes self-serve pricing, offers a 14-day trial, and refunds $50 for each confirmed numeric hallucination — the most transparent commercial model in our coverage.

The platform reflects a very small company. Company-filled sources put the team at roughly five to eleven people, and the only financing located is about $1.2m from 2021. Hudson Labs describes its controls as aligned with SOC 2 rather than attested, and we did not locate SSO or private deployment. Content is focused on US filings and transcripts; expert calls and broker research are not offered.

We lower the rating to Emerging, at 6.00, with a Stable outlook. This Rating Review applies the full sub-criteria standard used for every company covered since October and replaces the 6.70 published on 20 October. Under that standard, controls not documented publicly score low until located, and testimonials without measured outcomes do not count as return-on-investment evidence. The largest move is in Enterprise Readiness, now 4.71: controls described as aligned with SOC 2 are not an attestation, and SSO, private deployment and MNPI handling were not located. Data & Content, now 4.58, reflects content the company itself says it does not offer. Commercial Value, at 7.50, remains its strongest dimension and among the highest in our coverage. We located no material news since September.

Strengths

• Forensic depth. Fraud-likelihood scores across accounting, related parties, internal controls and governance, built by a CPA founder and refined since 2021; its findings have repeatedly been cited by financial press.
• Commercial transparency. Published Core pricing, a 14-day trial, an indicative Institutional price and per-project Deep Dive pricing.
• Accountability on numbers. A $50 refund for each confirmed numeric hallucination, restatement-adjusted five-year history and verbatim cited excerpts.
• Contractual no-training clause. The institutional terms prohibit using customer content to train, fine-tune or improve any model, with published retention periods and ten named subprocessors.
• Shipping cadence. At least nine dated launches between March and August 2026, including Market Intelligence, Agents, Deep Dive, Teams delivery and MCP in Claude.
• Applied-AI leadership. The CTO wrote O'Reilly's Designing Large Language Model Applications and worked on BigScience BLOOM.
• Independent accuracy signal. A third-party review rated multi-period extraction accuracy high.

Cautions

• Scale and capital. Company-filled sources show roughly five to eleven staff, and the only financing located is about $1.2m from 2021.
• Security attestation. Hudson Labs describes its controls as aligned with SOC 2; we did not locate an attestation, ISO 27001, a trust centre, SSO/SAML/SCIM or an SLA.
• Content scope. Coverage centres on US EDGAR filings, transcripts and decks with S&P consensus.
• Internal data. Document upload was unsupported in January 2026 and connectors were in beta or coming soon in June; buyers should confirm general availability.
• Tier gating. Forensic risk, export, sharing and MCP sit in the quote-only Institutional tier, so self-serve pricing does not reflect a full deployment.
• Model provider. Query content is sent to OpenAI; a zero-data-retention arrangement and MNPI handling were not located.
• Inconsistent figures. Client AUM, Core price and penetration-test cadence differ across the company's own pages.
• Evaluation evidence. Accuracy claims are vendor-run; no third-party benchmark score is published, and the refund guarantee excludes calculations and interpretation.