Product and intended use
Define the configuration, people, information and tasks being assessed.
Plans and what is in each
Public claim only
The pricing page (read 2026-09-27) publishes two plans. Free: $0 a month, 500 credits a month, responses to 4,000 words, default templates, three cross-checking models, access to the stated 45 million case and statute database. Paid: $110 a month after a 14-day free trial, billed monthly, unlimited prompts, responses to 10,000 words, premium templates, 24/7 technical support, all three functions, 'Multi-step AI agents for better accuracy' and six jurisdictions. The paid plan is titled 'Enterprise' on the card and 'PROFESSIONAL' in the detail list beneath, and the comparison block states '$50 - $110/mo'.
Ask: How many plans exist, what are their exact names and prices in Australian dollars, and what is the $50 figure in the comparison block?
What the product does
Public claim only
Three functions are published. Legal research returns a jurisdiction-specific answer with clickable citations the vendor states link to the actual case or statute rather than a summary. Contract drafting produces first drafts the vendor describes as jurisdiction-aware with governing law and compliance clauses included. Document review takes an uploaded document, flags risky clauses by severity and identifies missing protections. The vendor describes a verification layer it calls Osiris in three stages: the query is matched only against its database; three separate AI models then check the proposed answer against the source material without seeing each other's outputs, and any disagreement means the answer is flagged and not shown; then a cited answer is returned. Named features also include 'Access to CaseBase' and default or premium templates.
Ask: Demonstrate the Osiris flagging path on a question where the database has no answer, and show what the user sees when the three models disagree.
Jurisdictions covered
Public claim only
The vendor states coverage of six jurisdictions: Australia, New Zealand, the United Kingdom, the United States, Canada and India. The security page maps each to a privacy framework, listing Australia against the 'Privacy Act 1988 (reformed)' and stating only that the product 'is built to meet each one - not a single, watered-down global standard'. Court coverage, currency and how Australian state and federal material is handled are not published.
Ask: Which Australian courts, tribunals and consolidated legislation are in the database, to what date is each current, and how often is it refreshed?
Release stage and scale
Public claim only
The about page states the first prototype was built in 2023 during a law school exchange in Munich, that about 50 solo practitioners in Australia and New Zealand were beta testers in 2024, that the product launched publicly on Product Hunt in February 2025 with more than 500 signups in the first week, that the Osiris verification layer shipped in late 2025, and that more than 2,500 lawyers across six jurisdictions use it in 2026. The same figure, '2.5k+', appears on the security and pricing pages. None of these figures is independently sourced.
Ask: How many paying Australian firms and users are on the product today, and can you provide reference customers we may contact?
Vendor content addressed to AI systems, and two page observations
Public claim only
Checked on 2026-09-27: a URL map of evatt.ai returned 31 URLs including the full blog with no /llms.txt, /ai-info or /for-ai path, and the six pages read contain no block addressed to AI models, assistants or crawlers and no instruction on how the product should be described. Two dated page observations are recorded because a firm doing due diligence would want them, with no conclusion drawn from either. First, as at 2026-09-27 the security page displays a customer quotation attributed to 'Zoomerr' whose attribution line beneath reads 'Author NameRole, Firm', which is unfilled placeholder text. Second, the pricing page carries a competitor comparison block, sourced to 'Competitor pricing from public sources and verified user reports - Q1 2025', which sets out other vendors' prices and buying processes.
Ask: What is the source and date of each customer quotation on the security page, and of each competitor figure in the pricing comparison block?
Data, controls and requirements
Compare the actual configuration with firm, client and jurisdiction requirements, supported by evidence.
Uploading client documents
Not confirmed
Document review works by uploading a document, so client material leaves the firm. The security page asserts the product is aligned with 'the highest legal-ethical standards' and poses the question 'Could a disclosure to Evatt waive attorney-client privilege?', answering it with per-firm isolation, zero retention and no training. Neither the terms and conditions nor the privacy policy read on 2026-09-27 contains any confidentiality undertaking, data-processing clause, privilege provision or restriction on what may be uploaded, and there is no DPA. The privacy policy addresses only personal information collected through the website and enquiry forms.
Ask: Will you sign a confidentiality and data-processing agreement covering uploaded client documents and prompts, and where is the privilege and confidentiality position stated contractually rather than on the security page?
Foundation model providers
Not confirmed
No foundation-model provider is named on any of the six pages read. The vendor refers repeatedly to '3x AI models', 'Powered by 3x AI language models' and 'Three separate AI models check the proposed answer against the source material', and to a verification layer it calls Osiris, without identifying any model, provider, version or hosting arrangement. No customer model choice is described.
Ask: Name each of the three models, its provider, the entity you contract with, and the country in which each model endpoint runs.
Subprocessors
Not confirmed
No subprocessor list exists on any page read. The privacy policy names categories only: 'Hosting providers, Payment processors, Email or analytics service providers', and states such third parties receive information 'as long as they agree to keep this information confidential'. The three AI models are not identified, so the model providers are also unnamed. The pricing page mentions payment by major credit cards or a 'Link wallet' without naming the payment processor.
Ask: Provide a dated subprocessor list with entity, country and purpose, covering model providers, hosting, payments, email and analytics.
Training on customer content
Public claim only
The statement is unambiguous but appears only on marketing pages. The security page states 'No model training on client data. Not your prompts, your documents, or your outputs. Evatt never trains on client data - full stop.' and lists 'Train on your data' under 'What we never do'; the pricing page footer line states 'Your data is never used for training'. Neither the terms and conditions nor the privacy policy read on 2026-09-27 mentions model training, prompts or uploaded documents at all, and there is no DPA to carry the commitment.
Ask: Put the no-training commitment into the agreement, extended to the three model providers and to any fine-tuning, evaluation or prompt-improvement use, and confirm who can vary it.
Feedback and usage data
Not confirmed
No feedback control, rating mechanism or telemetry statement for the product was found on any page read. The privacy policy says personal information may be used to 'Improve Our Software and Website - Gather feedback to enhance user experience and functionality', which is about website and enquiry information rather than queries or documents, and describes cookies used for preferences and analytics. Nothing states whether prompts, answers or usage events are retained for product improvement, which sits in tension with the zero-retention statement on the security page.
Ask: What usage data, prompts or outputs do you retain for product improvement or debugging, for how long, and can a firm switch that off?
Retention of queries and documents
Public claim only
The security page states 'Zero data retention - Queries and documents aren't retained after your session. No logs kept beyond your control.' and 'Retain beyond your session' appears under 'What we never do'; the FAQ adds that privilege is maintained 'by erasing your data as soon as you log out'. No period, no backup position and no exception list is given. Neither the terms and conditions nor the privacy policy read on 2026-09-27 mentions retention of queries or documents. The paid plan advertises persistent-sounding features such as premium template access, and the Free plan a monthly credit allowance, without explaining what is stored to support them.
Ask: Define 'session' and 'zero retention' precisely: what is written to disk or to logs during a session, what survives logout, and what is in backups?
Deletion and erasure
Not confirmed
No deletion process is published. The security page relies on session purge at logout rather than a deletion right, and the privacy policy contains no data-subject request process, no erasure right and no timeframe, offering only a support email address as a contact point. Account closure is described only as cancelling a subscription, with no statement about what happens to stored data.
Ask: How does a firm request deletion of its account, uploaded documents and query history, within what period, and will you certify deletion in writing?
Vendor access to content
Not confirmed
The security page raises 'Who has access to your data?' as one of three questions it says firms are right to ask, and answers it with architecture claims: 'Each firm's data lives in its own isolated environment - no shared indices, no cross-tenant leakage', and 'Share across firms' and 'Sell or expose your data' under 'What we never do'. No page read states whether vendor personnel can read queries or uploaded documents, under what approval, or whether such access is logged.
Ask: Can any of your staff or contractors read a firm's queries or uploaded documents, under what approval, and is that access logged and available to the firm?
Access control and administration
Not confirmed
Nothing on the pages read describes single sign-on, multi-factor authentication, roles, seat administration or an audit log. Access is by login at evattai.com; the vendor states no credit card is needed to start the Free plan, that setup takes 'under 4 minutes', and that there is no seat minimum, with team purchases arranged by email. The security page uses the phrase 'object-level security' without defining it.
Ask: Does the product support single sign-on, enforced multi-factor authentication, firm-level administration and an exportable audit log of who ran which query?
Encryption
Public claim only
The security page states 'Every query and document is encrypted in transit and at rest with AES-256-GCM. Nothing moves or sits in the clear.' and lists 'AES-256-GCM' among its standards. No key-management arrangement, customer-managed key option or protocol version for transport is described, and the claim appears only on a marketing page with no report or certificate behind it.
Ask: Who holds and rotates the encryption keys, are customer-managed keys available, and which independent report evidences the AES-256-GCM implementation?
Independent security assurance
Public claim only
The security page lists four items under 'Certifications & standards' and states the vendor is 'clear about what's certified versus in progress', without labelling which is which: 'SOC 2 Type II - Security, availability, and confidentiality controls, independently audited', 'AES-256-GCM', 'GDPR compliant', and 'ISO/IEC 27001 - Information-security management aligned to the standard'. 'Aligned to the standard' is not a certification claim. No auditor, report period, certificate number or scope is named for the SOC 2 Type II item, and security documentation is available only through a meeting-booking link. The FAQ separately refers to 'complying with SOC2 Type 2 data encryption'. SOC 2: the vendor's own page states "and never shared. - Zero data retention. - No model training. - Privilege-first infrastructure. - SOC 2 compliant. The Real Cost ## What Is Manual Research _Actually Costing Your Fi" (https://evatt.ai/, checked 2026-10-10).
Ask: Provide the SOC 2 Type II report with auditor, period and scope, and state plainly whether ISO/IEC 27001 certification has been granted, is in progress, or is an alignment claim only.
Contract terms, entity and forum
Public claim only
The terms and conditions read on 2026-09-27 carry no version or date. Their six clauses cover cookies, a licence limiting use of content to 'personal, non-commercial reference only', hyperlinking permissions, an 'as is' disclaimer of warranties as to completeness, reliability and availability, changes effective immediately on posting, and a contact block publishing 'Evatt AI (By Dawnview AI) ABN 79155289858', an Australian mobile number and 'Australia (GMT +8 AWST)'. There is no data-processing, confidentiality, security, retention, service-level, indemnity or liability clause, no governing law or jurisdiction clause, and no subscription or payment terms, despite the pricing page selling a monthly subscription. The site footer names Dawnview AI Pty Ltd; no ACN is published.
Ask: Provide a subscription agreement and DPA with governing law, confidentiality, security obligations, liability and termination, confirm the full contracting entity name and its ACN, and state whether the licence clause limiting use to 'personal, non-commercial reference' is intended to apply to fee-earning legal work.
Hosting region and cross-border transfers
Not confirmed
No hosting region, data-centre location or processing country is stated anywhere on the six pages read, including the security page. The privacy policy states 'Although we operate from Australia, we proudly serve clients in the United States, the United Kingdom, India, Canada, and New Zealand', and names hosting providers only as a category of third party. There is no Australian Privacy Principle 8 statement, no standard contractual clauses and no transfer-safeguard description, and the security page's jurisdiction table asserts the product 'is built to meet' each framework without naming a location.
Ask: In which country are queries and uploaded documents stored and processed, in which country does each of the three models run, and is an Australia-only option available?
Incident notification
Not confirmed
No breach-notification commitment, period or incident history was found in the terms and conditions, the privacy policy or the security page. The privacy policy states that no method of electronic storage or transmission 'can be guaranteed to be 100% secure'. The security page's jurisdiction table lists 'mandatory breach notification' as part of the Australian and New Zealand frameworks the product is said to be built to meet, which is a description of the law rather than an undertaking to the customer.
Ask: What is your contractual notification period from discovery of an incident affecting our documents or queries, including an incident at a hosting or model provider?
Change notice
Public claim only
The terms state: 'Evatt AI reserves the right to modify these Terms at any time. Any changes become effective immediately upon posting. If we make significant changes, we will post a notice on our website or otherwise inform you.' The privacy policy contains the same mechanism for itself. Neither document is dated or versioned on the page read. Nothing constrains changes to the three unnamed models, to the Osiris layer or to the database, and no notice period applies to them.
Ask: What advance written notice will we get before the models, the verification layer, the database or the terms change, and will you maintain a dated version history?
Export and exit
Not confirmed
No export capability is described for research answers, drafts, document-review findings or query history, and no exit provision appears in the terms. Cancellation is described on the pricing page as available at any time, including during the 14-day trial, with the FAQ adding that downgrading is done by cancelling the subscription. Nothing states what happens to a firm's data on cancellation, beyond the general session-purge claim on the security page.
Ask: On cancellation, what can we export, in what format, and what is deleted and when?
Connectors and integrations
Not confirmed
No integration with Microsoft Word, Outlook, a document-management system or a practice-management system is described on any page read. Document review is performed by uploading a file into the web application, and the vendor presents the product as 'All three in one workspace. No switching tools, databases or subscriptions.' No application programming interface is advertised. The terms include a section on third parties linking to the website, which is unrelated to product integration.
Ask: Are there any Word, Outlook, document-management or practice-management integrations or an API, and what data would each transfer?
Open-web access
Public claim only
The product page states: 'Your question is matched exclusively against our database of cases and statutes - never the open internet, never Wikipedia, never a legal blog', and the FAQ states the product's 'training is restricted to a verified database of 45 million+ cases and statutes'. The exclusion is asserted on marketing pages only and the database itself is not described in verifiable terms, beyond the retrieval-source list naming Westlaw, LexisNexis, AustLII, CanLII and UK Legislation.
Ask: Does any part of the product issue an outbound internet request at query time, including for citation links, and can you evidence the closed-corpus claim?
Agent actions and confirmation
Public claim only
The paid plan lists 'Multi-step AI agents for better accuracy' on the pricing page, with no further description of what those agents do, what tools they can call or whether any step runs without the user. Everything else described on the pages read returns content to the user inside the application: answers with citations, drafts, and clause flags by severity. Nothing describes the product sending email, filing, or writing to any external system.
Ask: What do the multi-step agents actually do, which steps run without a user action, and can any of them reach outside the session?
Human review before use
Public claim only
The vendor states every citation is clickable and traceable to the primary source, that 'If any model cannot find a supporting source, the answer is flagged - not published', and that the product 'Flags uncertainty - never presents unsupported claims as fact'. Its FAQ states the product cannot replace lawyers. No product control requiring a named reviewer, sign-off or recorded verification before an output is used is described, and the terms contain no clause allocating responsibility for checking outputs.
Ask: What in-product control records that a lawyer has opened the cited authority before an answer is relied on, and can that be made mandatory for a firm?