Trust and Security
Elefant is a legal AI platform built by Elegant Elephant Pte Ltd d.b.a. Elefant, a Singapore company, for law firms and in-house legal teams in Singapore, Malaysia and other Commonwealth and Southeast Asian jurisdictions. This page sets out where your data lives, who can see it, how it is used and what we commit to in writing. If a question is not answered here, email us and we will answer it.
Data hosting and residency
| Your data | Where it is held or processed |
|---|---|
| Account details, uploaded documents and chats | Google Cloud, Osaka, Japan |
| Database backups | Google Cloud, a multi-region location in Asia. Automated backups are kept for 14 days |
| Search questions | Elastic, on Google Cloud, Iowa, United States. The search index holds public legal sources, not your files |
| Scanned PDFs, while the text is read | Google Document AI, Singapore |
| Requests to AI models | Pydantic's AI Gateway, United States, then the model provider. See Subprocessors |
If you need your data held in a different region, write to us. There is no region setting in the product today; we plan to add one. We do not operate our own data centres; physical security is covered by Google Cloud's certifications.
How your data is used
- Training. We never use your documents, prompts or outputs to train or improve any model, ours or a third party's, on any plan. There is no exception. Elefant runs no model training at all. Our model providers' terms are set out under "AI models and providers" below.
- Search quality. Your settings include a preference called "data training". It is on by default, on every plan, and you can switch it off at any time; only you can change it. With it on, searches that returned nothing may be used to improve our search quality tests, not any model.
- Retention. Uploaded documents and conversation history are kept until you delete them, or your account is closed. There is no shorter automatic retention period to set. Short-lived processing data (search caches, draft sessions, download links) expires on its own, within at most 7 days.
- Deletion. You can delete individual documents, files, chats and briefcases (matters), and your saved memory, at any time; each is deleted outright. When you close your account it is closed at once: your sessions end and no one can sign in to it. Our policy is to keep a closed account for a seven-day recovery window, during which support can restore it at your request, and then delete the account and its data. Deletion is carried out by our team; if you need it completed by a particular date, tell support and we confirm when it is done. Once deleted, copies age out of our automated backups within 14 days. If you were a member of an organisation's account, documents and matters you contributed stay with that organisation; your name is removed from its records.
- Memory and analytics. Elefant's memory of your preferences and its server-side usage analytics are each on by default, on every plan, and can be switched off by the user, each independently, in settings. Memory can also be turned off for a single chat.
AI models and providers
Elefant combines third-party foundation models with our own retrieval pipeline over primary-source legal data covering 16 jurisdictions. We use three categories of AI processing: foundation models that answer, draft and review; a smaller model that rewrites your search so it finds the right sources; and optical character recognition for scanned documents. We change models as better ones become available, so this page does not name them; the companies that may process your content are listed under Subprocessors below and at elefant.legal/subprocessors.
Our guarantee is the same whichever model answers: no provider that receives your content may train on it. Only providers whose API terms we have recorded as no-training are allowed to receive client content, and that rule is enforced in code and covered by tests, not left to configuration. Requests reach the providers through an AI gateway in the United States. Some providers keep requests for up to 30 days to check for abuse.
Outputs are grounded in retrieved primary sources, and our citation check looks up each cited authority in our corpus so users can see which were found and which were not. The check reduces, but does not remove, the need for a lawyer to check the work.
Security controls
What is listed as in place is in place today; what is not yet true is marked planned.
- Encryption. Your data is encrypted at rest and in transit. Customer-managed keys are not offered.
- Access. Production access is controlled in Google Cloud under a written quarterly access-review procedure. The first baseline review is planned. Access by Elefant's platform administrators is logged: a support sign-in shows account state and settings while document and chat content is withheld, and every such session is logged. Direct infrastructure access is limited to named administrators under the quarterly access review.
- Authentication. Sign-in is by email and password (at least 12 characters), emailed link or code, Google or Microsoft account, or passkey. Two-factor authentication with an authenticator app and backup codes is available to every user. Server-side enforcement of two-factor for administrators and for organisations that require it is planned. Single sign-on: an organisation administrator can register an OpenID Connect or SAML 2.0 identity provider for a work-email domain in organisation settings, and members then sign in with their work email. Organisation roles (owner, admin, counsel, staff, member, external) control who can manage members, billing and settings. A briefcase (matter) is private to its owner unless shared, with view, comment or full access.
- Logging. Sign-ins and account changes, sharing, API key and integration changes, and changes to documents, briefcases and chats are logged. Security and audit records, such as successful sign-ins and administrative actions, are kept for the life of your organisation's account and deleted with it; there is no shorter retention period. They record who did what and when, including email addresses, names and approximate IP addresses, but not the text of your documents or chats. Application logs are scrubbed of e-mail addresses, bearer tokens, API keys and Singapore NRIC/FIN numbers before they are written. Organisation administrators can see their organisation's audit history in-product (sign-ins, API keys, sharing, exports and downloads, document and matter changes), filter it and download it as a CSV report, and each user can see their own activity, including sign-ins. A full extract that also covers the actions of Elefant's own platform administrators on your account (support sign-ins, account changes) is available on request.
- Tenant isolation. Each organisation's data is separated in the database by row-level security keyed to the organisation and enforced on every request, so a request without an organisation is refused rather than answered. Your documents are stored in our database, not in the shared legal search index.
- Network. Application services run on Google Cloud; access to them is restricted and monitored.
- API and MCP server. The API checks each key against the scopes it was given. API keys can be limited to named scopes, given an expiry date, and revoked by the user who created them or by an organisation administrator; organisation integration keys are created and revoked by administrators. The MCP server authenticates on its own path, by OAuth or a key; accepted risks and their mitigations are listed in our security policy.
- Backups and recovery. Automated database backups are kept for 14 days.
- Dependency and container scanning. Every backend deployment and every merge runs automated dependency and container vulnerability scans; findings are triaged under our security update policy (critical within 48 hours, high within a week). GitHub dependency alerts, automatic security updates and secret scanning with push protection are on.
Compliance and certifications
Elefant is working towards ISO/IEC 27001 and SOC 2. Neither is yet certified; we will not describe ourselves as certified until an independent auditor says so. Our information security policies, risk register and control evidence are available to prospective customers under NDA. We comply with the Singapore Personal Data Protection Act 2012. For Malaysian customers we support compliance with the PDPA 2010 as amended.
Subprocessors
We use subprocessors to deliver the service. The list, with what each does and where it processes data, is at elefant.legal/subprocessors.
Models whose terms do not meet our no-training bar are not offered.
Sub-processor additions or replacements are posted at elefant.legal/subprocessors at least 14 days before the change. If a change materially reduces the protection of your content, you may terminate within that period.
elefant.legal/subprocessorsConfidentiality and legal privilege
A general standard is beginning to take shape for lawyers who use AI: the tool may help, but the duties to protect client confidences and to check the work stay with the lawyer. Singapore is one example. A legal practitioner there must not knowingly disclose information confidential to a client (Legal Profession (Professional Conduct) Rules 2015, rule 6(2)), and the High Court has said that lawyers bear ultimate responsibility for verifying what they place before the court (Tan Hai Peng Micheal v Tan Cheong Joo [2026] SGHC 49 at [32]). The High Court of England and Wales has said that lawyers who use AI for legal research have a professional duty to check its accuracy, and that judicial guidance on protecting confidentiality when using AI is as relevant to lawyers as to judges (R (Ayinde) v London Borough of Haringey [2025] EWHC 1383 (Admin) at [7], [15]).
Elefant is designed to support that duty, which remains with the people who carry it. Your content:
- is confidential to your organisation;
- is never used to train any model;
- is sent only to providers whose terms we have recorded as no-training, through Pydantic's AI Gateway (see Subprocessors);
- is access-logged; and
- is deletable on your instruction.
Whether legal professional privilege is preserved depends on how your firm uses the tool and on the applicable law, not on the software itself. We cannot advise you on that question; the contractual commitments and technical facts on this page are offered to inform your own assessment and client disclosures.
Incident response
Our incident response process covers detection, triage, containment, notification and post-incident review; a one-page summary will be published alongside this page. If we become aware of a security incident affecting your data, we notify your nominated contact with the information you need to meet your own obligations. We do so without undue delay and, under a DPA, within three calendar days of becoming aware. We notify the Personal Data Protection Commission where the Personal Data Protection Act 2012 requires it, as soon as practicable and no later than three calendar days after we assess that a breach is notifiable (ss 26C, 26D), and affected individuals as the Act requires; where the data is yours to answer for, we give you what you need to make your own notification. A written root-cause report follows once the incident is closed. Availability incidents are posted on status.elefant.legal.
Business continuity
Automated database backups are kept for 14 days with point-in-time recovery over the last seven days. A restore from backup onto a fresh instance was tested on 4 November 2026 and completed in about ten minutes. There is no uptime commitment unless an Order Form states one. You can download your own documents, chat history and account information at any time as a ZIP from your settings, and export single chats, documents and briefcases. An export of a whole organisation, and escrow of your data so that it survives Elefant, are in progress.
Penetration testing and vulnerability disclosure
Not yet. An external penetration test of the web application, API and MCP server is scheduled; its summary will be available under NDA once complete. Automated dependency and container scanning runs on every backend deployment and at the merge gate, as reports. Our security update policy is to patch critical vulnerabilities within 48 hours and high within a week. Researchers can report vulnerabilities to [email protected]; we acknowledge within 24 hours, ask for up to 90 days before public disclosure, and do not pursue good-faith researchers. A security.txt and a disclosure page are planned.
Contracts
Enterprise customers contract on a Master Services Agreement, a Data Processing Agreement reflecting PDPA requirements (and GDPR terms where relevant) and a confidentiality undertaking. The DPA covers instructions, subprocessors, security measures, breach notification, data subject requests, deletion on termination and audit rights.
Contact
Security and privacy questions: [email protected]. Elegant Elephant Pte Ltd, UEN 202332509W, 7 Temasek Boulevard, #12-07, Suntec Tower One, Singapore 038987.
Privacy PolicyQuestions buyers ask
One line each; the detail is in the section linked.
- Is our data used to train AI models? No, never, on any plan. See How your data is used and AI models and providers.
- Where is our data stored? Osaka, Japan; search queries reach Iowa, United States; scanned PDFs are read in Singapore. Write to us for a different residency. See Data hosting and residency.
- Are you ISO 27001 or SOC 2 certified? Not yet; in progress. See Compliance and certifications.
- Who at Elefant can see our documents? Platform administrators, under logged access. See Security controls, "Access".
- Can we delete our data, and how fast? Yes, item by item at any time. A closed account is frozen at once and, under our policy, deleted after a seven-day recovery window; backups age out within 14 days after that. See How your data is used, "Deletion".
- Will you sign our DPA and complete our security questionnaire? Yes to a PDPA-aligned DPA and standard questionnaires; a penetration-test summary follows the first test. See Contracts and Penetration testing and vulnerability disclosure.