Legal Tech / Knowledge & document management
Hebbia Matrix
Hebbia, Inc. · HQ US · founded 2020 · Last checked 2026-09-22
Last checked = the date we last re-opened the vendor's published pages. Facts are what the vendor documents, not what it claims in marketing.
Agentic document-analysis platform ('Matrix') that runs natural-language queries across large document sets with cited answers; used by law firms for litigation and transactional work.
- Category
- Knowledge & document management · document analysis, multi-document Q&A, litigation support, transactional
- Built for
- Large firms, In-house legal
- Model providers (disclosed)
- OpenAI, Anthropic, Google, Microsoft
- Data residency options
- US, EU
- Pricing
- enterprise
- Markets
- US, EU
- Last checked
- 22 Sept 2026
What the vendor documents
Recorded from the vendor's own trust, privacy and terms pages. "Not disclosed" means we found no public statement, not that the control is absent.
What the public documentation states
Hebbia’s public pages describe Matrix as an agentic document-analysis product for querying large document sets with cited answers, including law-firm litigation and transactional work. The DPA dated 27 April 2026 publishes subprocessors and names Google, OpenAI, Microsoft, Anthropic, Cerebras, Baseten, Groq, Fireworks and Modal for AI-related processing. The privacy policy dated 8 May 2026 says Hebbia does not use personal data in customer data to train or improve generalized or third-party AI/ML models. It also says data processing is restricted to authorised geographic regions set in the DPA or customer agreements, and service data is retained for the customer relationship and a period after. The security and homepage text states ISO/IEC 42001:2023, SOC 2 Type 2 and ISO/IEC 27001:2022, plus data encrypted at rest with AES 256. The privacy policy says Hebbia does not use personal data contained in customer data to train or improve generalized or third-party AI/ML models. The pricing page states Enterprise customization but no price. The security page states a “never hallucinates” claim; no method was published.
In the vendor's words
“never hallucinates and gives correct, verifiable answers”
“never trains on your data”
“do not use Personal Data contained in Customer Data to train or improve generalized”
“restricted to the authorized geographic regions set forth in our DPA”
Not disclosed on public pages
"Not disclosed" records that no public statement was found where we looked, on the date shown. It does not assert that the control, term or figure is absent.
On pages read on 2026-09-22, Hebbia did not publish a free trial, list price, minimum term, seat minimum, accuracy method, private-deployment option or zero-retention commitment. A law-firm procurement or DPIA process normally needs these to map matter-data hosting, processor obligations, client confidentiality, and supervision of AI answers.
Questions to put to the vendor
- Which AI subprocessors from the 27 April 2026 DPA handle client matter data for each Matrix workflow?
- What geographic regions are authorised for this customer, and how are those regions recorded in the DPA?
- What benchmark or test method supports the security page's never-hallucinates statement?
- Can Hebbia provide ISO 27001 and SOC 2 Type II report details, including auditor and current report period?
- Does any model provider retain prompts, outputs or embeddings, and where is zero retention documented?
Accuracy evidence
No published accuracy or hallucination evaluation found as of 2026-09-22.
Related incidents
No incidents are recorded for this tool as of 2026-09-22.
Ownership · >$160M raised; $130M Series B led by a16z (2024); investors Index, GV, Peter Thiel.
Sources
Report an issue with this record →
Cite this record
SafeLegalAI Legal Tech Tools, "Hebbia Matrix", safelegalai.com/tools/hebbia (last checked 2026-09-22, accessed 2026-10-07). Data: CC BY 4.0.
What the vendor has announced, dated
- Ice Miller uses Hebbia across the firm · Named deployment
- Daloopa is available in Hebbia · Integration
- Hebbia publishes September Disclosure updates · Feature release
- Hebbia introduces Matrix 2.0 · Product launch
- Hebbia introduces Max · Product launch
- Hebbia discusses token efficiency at scale · Research or benchmark
- Hebbia adds single view for data integrations · Feature release
- Hebbia integrates Snowflake · Integration
- Hebbia publishes July Disclosure updates · Feature release
- Hebbia publishes June Disclosure updates · Feature release
First-party statements from the vendor's own pages, as dated; "announced" is not "in use today". This month across all vendors · release timeline.
Documented facts, side by side
- Hebbia Matrix and DeepJudge
- Hebbia Matrix and iManage AI (Ask iManage)
- Hebbia Matrix and NetDocuments ndMAX
Facts from both vendors' pages in one table; never a ranking. All comparisons.
Also in Knowledge & document management
SafeLegalAI is a research publication, not a law firm. Nothing here is legal advice, and no lawyer–client relationship arises from reading it. Rules change; always check the official document linked on each record and take advice on your own situation. Researched and drafted with AI assistance; verified against primary sources and edited by SafeLegalAI. The linked official documents are the record — our summaries are not the law and are not legal advice.
Vendor, product, court and regulator names and logos are the property of their respective owners and are used solely to identify the product, body or document referred to; SafeLegalAI claims no rights in them. SafeLegalAI is not affiliated with, endorsed by, sponsored by or approved by any listed vendor, except where a record is labelled an affiliated product. Facts are recorded from vendors' published pages on the date shown; "not disclosed" means not found on the pages read, not that the vendor lacks it. No ranking, rating, review, recommendation or warranty is implied. Mark owners who prefer a text-only reference, and vendors seeking a correction, may write to the report page. Trade mark and quotation policy.