The version that applies to you is the one you accept in the room. This copy is for information. "Sealroom" is the working name of the Haihtuva service.
Sealroom AI Use Notice and Acceptance
Version 1.3 (final) Effective date: 12 October 2026
Summary
- AI output may be inaccurate or incomplete and may misread tables, images, document versions or citations.
- Users must verify material AI-generated conclusions against the source documents before relying on them.
- Customer Results are generated analytical materials. They do not by themselves constitute legal, financial, investment, regulatory or medical advice from BSUM.
- The approved provider, model, deployment, endpoint, processing location, retention and logging settings are recorded in the Room’s Security Specification.
- BSUM does not use Room Content or the substance of User questions to train or fine-tune general-purpose or customer-facing AI models. Provider processing is governed by the verified configuration and provider terms identified in the Security Specification.
- Sealroom is not GxP-validated, a medical device or a clinical decision-support system unless expressly agreed in writing.
- BSUM’s non-public prompts, system instructions, analytical workflows, taxonomies and configuration logic are protected Platform Information.
- This notice forms part of the Sealroom Terms of Service (the "Terms") between Biotech Start-Up Management Oy ("BSUM") and the Customer. The numbered provisions prevail over this summary.
1. Nature and limitations of AI output
1.1 Generated output
Reports, answers, briefs, findings, drafts and other Customer Results may be generated or assisted by AI models.
Customer Results may contain:
- factual errors;
- interpretation errors;
- citation errors;
- calculation errors;
- extraction errors;
- translation errors;
- omissions;
- unsupported conclusions; or
- information not contained in the source materials.
An AI model may also:
- overlook relevant information;
- misread tables, images or scanned documents;
- confuse document versions;
- fail to identify a changed or superseded provision;
- omit qualifications or exceptions; or
- attribute information to the wrong source.
1.2 Verification
Users must verify material Customer Results against the source materials before relying on them.
Citations and page references assist verification but do not guarantee that:
- the reference is accurate;
- the source supports the generated conclusion;
- all relevant sources were identified; or
- the cited text remains current or complete.
1.3 Professional responsibility
Customer Results are generated analytical materials and do not by themselves constitute legal, financial, investment, tax, regulatory or medical advice from BSUM.
Customer Results:
- do not replace advice from appropriately qualified professionals;
- do not constitute a recommendation by BSUM to complete or reject a Transaction;
- do not constitute a recommendation concerning Transaction terms; and
- do not transfer responsibility for decisions to BSUM or the model provider.
1.4 Decisions concerning individuals
Customer Results must not be used as the sole basis for a decision based solely on automated processing that produces legal effects concerning an identifiable person or similarly significantly affects that person, except where:
- the processing is permitted by applicable law; and
- all required safeguards have been implemented.
Where a Customer Result informs a decision concerning an identifiable person, the Customer must:
- establish an appropriate legal basis;
- assess applicable automated-decision-making and profiling requirements;
- provide meaningful human review;
- verify the relevant source information;
- allow correction of materially inaccurate information where required; and
- implement any other safeguards required by applicable law.
1.5 Human responsibility
The Customer and its Users remain responsible for:
- deciding whether and how to use Customer Results;
- selecting appropriate reviewers;
- verifying material findings;
- assessing legal and regulatory requirements; and
- decisions taken on the basis of Customer Results.
1.6 Regulated life-science use
Unless expressly agreed in an Order, Sealroom and Customer Results:
- are not validated for GxP-regulated use;
- are not a medical device;
- are not a clinical decision-support system;
- are not designed to control a medical device or laboratory instrument;
- must not be used to make autonomous decisions concerning diagnosis, treatment, patient eligibility, clinical-trial participation or participant safety; and
- may be used for commercial, intellectual-property, regulatory, quality, scientific and transaction-document analysis subject to appropriate expert and human review.
Use in a regulated workflow does not by itself make Sealroom validated or compliant for that workflow.
1.7 Scientific and technical limitations
Customer Results may not identify all:
- scientific risks;
- technical risks;
- intellectual-property risks;
- regulatory risks;
- quality risks;
- clinical risks;
- cybersecurity risks; or
- commercial risks.
Users must apply suitably qualified subject-matter review appropriate to the Permitted Purpose.
1.8 Warranty and liability
BSUM does not warrant that Customer Results will be accurate, complete, current, error-free or suitable for a particular decision.
BSUM’s warranties and liability are governed by the Terms.
2. Room-specific AI configuration
2.1 Required approvals
Before Transaction Information or other protected Customer Content is submitted to a model, the configuration must be approved by each Customer or Room side whose approval is required under:
- the applicable Room roles;
- any Transaction NDA;
- the Order;
- applicable data-protection arrangements; or
- the Security Specification.
The Security Specification identifies the required approvers for each processing mode.
2.2 Security Specification
The Security Specification records the approved configuration and identifies, where relevant:
- model provider;
- model family;
- model or service version, where available;
- deployment identity;
- endpoint;
- processing region or data zone;
- account owner;
- processing mode;
- grounding sources;
- output destination;
- retention settings;
- logging settings;
- safety and abuse-monitoring settings;
- known human-review conditions;
- whether content leaves the Room’s confidential-computing boundary;
- applicable output rules;
- permitted document classes;
- applicable BSUM-provided instruction sets and their fingerprints; and
- the required approvers.
The Security Specification does not disclose protected prompts or other Platform Information merely because it records their fingerprint or identifier.
2.3 Changes
A material change to an approved configuration requires new approval before the changed configuration is used for protected Customer Content.
A change is material where it may reasonably affect:
- confidentiality;
- data protection;
- information security;
- data location;
- provider access;
- retention;
- human review;
- output behaviour;
- output permissions; or
- the agreed risk profile.
A change to any of the following may be material even where the commercial model name remains unchanged:
- underlying model version;
- model capabilities;
- provider processing terms;
- provider retention conditions;
- safety filtering;
- abuse monitoring; or
- deployment type.
New approval is not required for routine maintenance that does not materially change the approved configuration or risk profile.
2.4 Side-specific settings
AI settings that affect only one side’s private work may be separately identified in the Security Specification.
Examples include:
- a Seller preparing draft answers;
- a Buyer generating a private analysis;
- a Customer using a separately approved model under its own account; or
- a Customer restricting particular document classes from model processing.
Approval of one side’s private output does not grant the other side access to that output.
2.5 Processing outside the Room
Where an approved model runs outside the Room’s confidential-computing boundary, the information required for the request is transmitted to the endpoint identified in the Security Specification and processed under the applicable provider terms and configuration.
The Security Specification identifies:
- the destination;
- region or data zone;
- account owner;
- configured retention;
- logging;
- abuse monitoring;
- known review processes; and
- settings that cannot be disabled.
BSUM does not represent that a provider performs zero retention, zero logging or zero human review unless that conclusion has been verified against the applicable technical configuration and contractual terms.
Where BSUM selects or contracts with the provider, BSUM remains responsible for complying with its obligations concerning that provider under the Terms and applicable data-processing arrangements.
Where the Customer selects and controls the provider through its own account, the allocation of responsibilities is identified in the Security Specification.
3. Training, retention and operational data
3.1 No model training by BSUM
BSUM does not use Customer Content, Transaction Information, Customer Results or the substance of User questions to train or fine-tune a general-purpose or customer-facing AI model unless the Customer has separately and expressly instructed that use in writing.
Where BSUM selects the model provider, BSUM will use the available contractual and technical configuration intended to prevent the provider from using that information to train or fine-tune general-purpose foundation models.
Any provider-side retention, human review, abuse monitoring or service-improvement processing that cannot be disabled must be disclosed in the Security Specification before approval.
For models BSUM provides on Microsoft Azure, Microsoft's abuse monitoring for Azure AI services applies under Microsoft's terms unless Microsoft has approved an exemption for BSUM's deployment; the Security Specification states which applies to the Room's model. Models processed outside the EU/EEA (Claude models provided by BSUM are processed in the United States) are marked as such in the Security Specification and are used only after both sides approve it.
3.2 Operational telemetry and statistics
BSUM may use content-free telemetry and operational statistics to:
- operate the Service;
- maintain availability;
- prevent misuse;
- investigate security events;
- measure technical performance; and
- improve the Service.
Such telemetry and statistics must not contain:
- Transaction Information;
- Customer Results;
- document text;
- the substance of User questions; or
- private analytical work.
Where operational statistics remain personal data, BSUM processes them in accordance with its privacy notice and applicable law.
BSUM may use Feedback as permitted by the Terms.
3.3 Customer-controlled model accounts
A Customer that chooses to use a model through its own provider account is responsible for that account’s:
- provider agreement;
- data location;
- permissions;
- retention settings;
- logging settings;
- abuse-monitoring settings;
- access controls; and
- lawful processing.
The Security Specification identifies:
- that the account is Customer-controlled;
- the responsible Customer;
- the model and provider;
- responsibility for retention and logging;
- the applicable provider terms; and
- whether and to what extent BSUM can verify the configuration.
BSUM does not guarantee settings that it cannot control or verify.
4. Customer Results and Platform Information
4.1 Customer Results
Section 10 of the Terms governs Customer Results.
The Customer may use Customer Results as permitted by:
- the Terms;
- applicable law;
- the Permitted Purpose; and
- any applicable Transaction NDA.
No rights are transferred in Transaction Information, third-party material or Platform Information merely because that information is contained in or reflected by a Customer Result.
4.2 Protected Platform Information
The following non-public elements implemented or supplied by BSUM are Platform Information:
- prompts;
- system instructions;
- proprietary analytical workflows;
- proprietary taxonomies;
- configuration logic;
- output-control logic;
- reusable proprietary templates;
- evaluation methods;
- orchestration logic; and
- security rules.
Ordinary report headings, commonly used due-diligence categories and Customer Results delivered to the Customer are not Platform Information merely because Sealroom generated them.
Receipt of a Customer Result does not grant access to or rights in the underlying:
- prompts;
- instructions;
- taxonomies;
- configuration;
- workflow logic;
- methods; or
- security design.
4.3 Protection against extraction
Users must not attempt to extract, reveal, copy, reconstruct or systematically derive BSUM-provided prompts, system instructions or other Platform Information.
Users must not:
- infer protected instructions through systematic repeated testing;
- compare outputs systematically to derive hidden prompts or workflow logic;
- use prompt injection or comparable techniques for the purpose of revealing, changing, bypassing or reconstructing protected prompts, instructions, policies, controls or workflows;
- obtain policy, developer or system messages;
- reconstruct a proprietary taxonomy or analytical process; or
- cause an AI model to disclose information outside the User’s authorised access.
4.4 Competitive use
The Customer and its Users must not use:
- Platform Information;
- extracted or reconstructed BSUM instructions; or
- Customer Results in a manner designed to reveal or reproduce Platform Information
to train, reverse engineer, benchmark for product development, develop or improve a product or service competing with Sealroom.
This restriction does not prevent the Customer from internally evaluating the quality and suitability of Customer Results for its own Permitted Purpose, provided that the Customer does not:
- disclose Platform Information;
- conduct prohibited testing;
- reconstruct BSUM’s protected methods; or
- use the evaluation for competitive development.
The independent-development and mandatory-law provisions in the Terms apply.
4.5 Unexpected protected output
BSUM does not guarantee that an AI model will never generate content that resembles a protected instruction, system message or other Platform Information.
A User who encounters or suspects such output must:
- stop further attempts to obtain similar content;
- avoid retaining, using or disclosing it except as necessary to report the incident;
- preserve only the information reasonably necessary for reporting; and
- notify the Customer administrator and BSUM without undue delay.
4.6 Unauthorised information disclosure
A User who suspects that a model has revealed:
- Platform Information;
- another party’s confidential information;
- information outside that User’s authorised access; or
- personal data not relevant to the permitted request
must stop further attempts, avoid further use or disclosure and notify the Customer administrator and BSUM without undue delay.
5. General provisions
5.1 Permitted Purpose
Permitted Purpose has the meaning given in the Terms and is identified for each Room in the applicable Order or Security Specification.
5.2 Rooms without a Transaction NDA
Where no Transaction NDA applies, references in this notice to rights or restrictions under a Transaction NDA do not apply.
Access and use remain subject to the Terms, applicable Order, Permitted Purpose, Security Specification, applicable law and valid Room-specific confidentiality and use conditions.
5.3 No waiver through visibility
Observation of a feature, workflow, output format, analytical structure or system behaviour through authorised access does not:
- make the associated Platform Information public;
- waive its protected status;
- grant a licence; or
- grant the right to reproduce or use it outside the Permitted Purpose.
5.4 No residuals right
No residuals right or licence is granted.
Retention of Platform Information in unaided memory does not authorise its use or disclosure in breach of the Terms, the Room Rules or this notice.
This does not restrict general professional knowledge, experience and skills that do not disclose, reproduce or materially incorporate Platform Information.
5.5 Mandatory law
Nothing in this notice excludes or restricts an activity or right that cannot lawfully be excluded or restricted by agreement.
Information obtained through such an activity may be used only within the scope and for the purpose permitted by applicable mandatory law.
5.6 Order of precedence
The documents apply as follows:
- the Terms govern the overall agreement;
- an Order prevails only for the commercial matters it expressly addresses;
- the Platform and Room Use Rules govern operational conduct;
- this notice governs AI-specific use, limitations and responsibilities; and
- the Security Specification governs the approved technical configuration.
The Security Specification does not override or reduce the platform-protection, confidentiality, acceptable-use or intellectual-property restrictions in the Terms, the Room Rules or this notice.
5.7 Survival
The provisions concerning:
- Platform Information;
- prompt and workflow protection;
- competitive use;
- unexpected disclosure;
- incident reporting; and
- acceptance evidence
survive termination or loss of access to the extent stated in the Terms or required by their nature.
6. Acceptance
6.1 Separate acceptance
Each User accepts this notice electronically and separately from the Terms and the Platform and Room Use Rules, as described in section 15 of the Terms.
6.2 Acceptance record
The acceptance record identifies:
- the User;
- the Customer;
- document title;
- document version;
- date and time;
- a fingerprint of the exact text accepted; and
- acceptance by the authenticated User.
6.3 Applicable terms
Finnish law and the dispute provisions in the Terms apply.
The English text prevails over any translation.