AI equity research

Hudson Labs

Formerly Bedrock AI

The auditor who made AI read the fine print

In 2019 Kris Bennatti, a Toronto CPA turned data scientist, quit her job to teach software to read company filings like a forensic accountant. Today her company, Hudson Labs, covers more than 10,000 public companies, all US issuers and ADRs.

Spotlight Images: Hudson Labs, MaRS

The Hudson Labs Co-Analyst screening companies for supply bottlenecks, with quoted evidence from their earnings calls

10,000+public companies the Co-Analyst covers, all US issuers and ADRsCompany-reported, as of Oct 2026 · Source for 10,000+ public companies the Co-Analyst covers, all US issuers and ADRs
$600B+assets managed by its clientsCompany-reported, as of Oct 2026 · Source for $600B+ assets managed by its clients
$50refund to paying subscribers for each confirmed figure missing from the source documents, under its No Hallucination GuaranteeCompany-reported, as of Jun 2026 · Source for $50 refund to paying subscribers for each confirmed figure missing from the source documents, under its No Hallucination Guarantee

Founders: Kris Bennatti; Suhas Pai
HQ: Toronto, Canada
Founded: 2019
Sector: Fintech, investment research
Website: Visit Hudson Labs

Born Capital is an investor in Hudson Labs.

The story

Hidden on page 101

Kris Bennatti started her career auditing financial statements at KPMG in Canada. Later, doing data science for Hansell McLaughlin, a Toronto corporate governance advisory firm, she was a rare mix: a CPA who also wrote code. She published a short computational study of cannabis company disclosures, and a local regulator asked her to run a text analytics workshop. The request showed her how little anyone was doing with the words inside public filings. Damaging facts, she argued, often sat in plain view around page 101 of reports too long and dull for most readers to finish. In November 2019 she quit to build Bedrock AI.

Labels checked by hand

Suhas Pai, an NLP researcher who had spent more than five years at IBM, joined full time as CTO in 2020. The core product went live in April 2021. It combed SEC filings for product defects, regulatory trouble and the accounting and governance choices behind the numbers. To teach it what fraud looks like, the team used SEC fraud enforcement actions as labels, then checked every label by hand, which took more than a month. In a back-test on old filings of Sino-Forest, a Toronto-listed timber company that Ontario's securities regulator found in 2017 had overstated its assets and revenue, the models rated it high risk. In 2021 the founders also said they did not sell to the companies they scored, to avoid conflicts of interest.

A Co-Analyst answer on Visa’s full-year guidance, with the earnings-call passage it was traced to — Every Co-Analyst answer is built to trace back to the filing, release or call it came from.

Two years early

In the summer of 2021 Bedrock went through a startup accelerator, and four Canadian securities regulators picked it for a cross-border testing program. Then came its most public call. Hudson Labs says that in 2022 its models flagged related-party dealings at Super Micro Computer involving the chief executive's brother, and that its related-party screen placed the server maker among the 100 companies with the highest related-party risk. On August 28, 2024, a day after a short seller's report, Super Micro said its annual report would be late, and its shares fell 19% that day. Two days later Forbes contributor Noah Barsky wrote that AI had spotted the risks two years before the filing delay. That November, Bennatti walked CNBC's Fast Money through the signals.

Every number, sourced

By late 2023 the company was doing more than hunting for fraud. It summarized earnings calls, drafted auditable investment memos and ran news feeds. On December 7, 2023, Bedrock AI announced its new name, Hudson Labs, which nods to Hudson Bay and Toronto, and to the Hudson River and Wall Street. The Co-Analyst, publicly launched in September 2025, is now the core product. It draws on SEC filings, call transcripts, investor decks and press releases for all US issuers and ADRs, which Hudson Labs pre-processes and tags for its own retrieval system. Answers return figures, guidance and management commentary across many periods, and Hudson Labs says each one can be traced to the disclosure behind it. In 2026 the team added Deep Dive research agents, topic agents and a Claude connector.

Money on accuracy

In tests by Bennatti's team, leading general-purpose models gave a wrong financial number in roughly 30% of answers to questions spanning more than a few reporting periods. In June 2026 Hudson Labs put money on its own answers. A paying subscriber who finds a management-disclosed figure in a Co-Analyst answer that cannot be traced to the filings, releases or calls behind it gets a $50 refund once the Hudson Labs team confirms the error. Hudson Labs lists Fortune 100 companies and top-five law firms among its users and says its clients manage more than $600 billion. In a testimonial on the company's site, Tom Elrod, a partner at the law firm Kirby McInerney, says the product has markedly improved the process of identifying securities fraud.

Visit hudson-labs.com

AI at Hudson Labs

Hudson Labs builds purpose-specific models trained on financial documents, and it says its software has been fully powered by language models since its first release in 2021. It is made for equity research, where a number counts only if it can be traced to its source.

Extracts figures, guidance and management commentary, each cited to its source
Agents run earnings, research and portfolio monitoring on a schedule the user sets
Forensic risk scores rate the likelihood of fraud from related-party, accounting and off-balance-sheet signals

The founders

Kris Bennatti
CEO and Co-Founder
Bennatti trained as a chartered professional accountant and has taught Python at an AI boot camp. She studied accounting and economics at McGill and data analytics at Schulich.
Suhas Pai
CTO and Co-Founder
Pai, an NLP and machine learning researcher, was a staff software engineer at IBM. He co-chaired privacy work on BigScience's BLOOM model, has chaired the Toronto Machine Learning Summit and wrote the O'Reilly book “Designing Large Language Model Applications” (2025).

Milestones

2019Nov
Kris Bennatti quits her Toronto data science job to start Bedrock AI.
2021Jul
Goes through a startup accelerator; four Canadian securities regulators select it for a cross-border testing program.
2023Dec
Bedrock AI renames itself Hudson Labs.
2024Aug
A Forbes contributor credits Hudson Labs' AI with spotting Super Micro risks two years before its filing delay.
2025Sep
Publicly launches the Hudson Labs Co-Analyst for institutional equity research.
2026Jun
Introduces a No Hallucination Guarantee: a $50 refund to paying subscribers for each confirmed untraceable figure.