This Defence Contract Addendum (the Addendum) supplements the Terms of Service and the Acceptable Use Policy between Bynn Intelligence, Inc. (Bynn) and an Eligible Customer. Military Use of the Service is permitted only under this Addendum. It takes effect only when signed by authorised representatives of both parties, together with an Order Form that identifies the Named Keys, the permitted purposes and the Accountable Role. Publication of this form is not an offer, and Bynn may decline to enter into it with any person (Section 2.2). Capitalised terms not defined in this Addendum have the meanings given in the Terms of Service.
Section 1. Definitions
#- Accountable Role
- the role within the Customer's organisation, named in the Order Form, that is accountable for the Customer's use of the Service under this Addendum.
- Eligible Customer
- a government body or contractor that meets Section 2.1 and has signed this Addendum.
- Harm and Targeting Component
- the part of the Safety Check that assesses whether a request seeks to harm others, as described in Section 2, Section 3.2 and Section 9 of the Acceptable Use Policy, and, where enforced, refuses it.
- Military Use
- has the meaning given in Section 4.2 of the Terms of Service.
- Mission Instructions
- what the Customer states in a request about its side, objectives, constraints and rules of engagement.
- Named Keys
- the API Keys listed in the Order Form for use under this Addendum.
- Protected Persons
- civilians and all other persons protected under international humanitarian law, including the wounded, the sick, prisoners of war and persons hors de combat.
Section 2. Eligibility
#Who may sign. Only defence and security bodies of the member states of the North Atlantic Treaty Organization, the member states of the European Union, Australia, Japan, New Zealand and the Republic of Korea, and contractors working for those bodies on their programmes, may sign this Addendum.
Due diligence and discretion. Before signing, Bynn verifies the Customer's identity, ownership and intended end use, and screens it against sanctions and export control lists. Bynn may decline to enter into this Addendum with any person, in its sole discretion and without giving reasons.
Export control. Each party shall comply with the export control and sanctions laws that apply to it, including Regulation (EU) 2021/821 on dual-use items and, where relevant, the export control laws of the United States, and Section 18 of the Terms of Service. The Customer shall obtain every licence or authorisation its use requires, and shall not re-export the Service or give access to it to any person outside Section 2.1.
Change of status. The Customer shall notify Bynn immediately if it ceases to meet Section 2.1, if its ownership or control changes, or if it or any of its owners becomes subject to sanctions. Access under this Addendum then ends unless Bynn agrees otherwise in writing.
Section 3. Permitted use and Mission Instructions
#Purposes. The Customer may use the Service for Military Use only for the purposes stated in the Order Form, such as logistics, readiness, threat assessment, intelligence triage, planning support and force protection.
A neutral Model. The Model is trained to treat every country, people and group equally and holds no built-in side.
The side comes in the request. The Customer states its side, objectives, constraints and rules of engagement in each request, in the state or the instructions. The Model answers from that operator's point of view. The Customer is solely responsible for the accuracy and lawfulness of its Mission Instructions.
Section 4. Uses forbidden in every case
#Important: Forbidden in every case
Even under this Addendum, and whatever its Mission Instructions say, the Customer shall not use the Service:
in breach of international humanitarian law, including the principles of distinction, proportionality and precaution, or of the Customer's own rules of engagement;
to choose, rank or target people by nationality, ethnicity, race, religion, sex or any other protected characteristic;
to weigh the lives or safety of Protected Persons differently by their origin or group;
to target Protected Persons, or civilian objects protected under international humanitarian law;
for biological, chemical, nuclear or radiological weapons;
for surveillance of civilian populations without a lawful basis;
to help any person harm themselves, for which reason the Self-Harm Component is never switched off; or
Section 5. Human judgment and accountability
#A person decides. A responsible person shall authorise every use of force that the Service informs, exercising appropriate levels of human judgment, consistent with United States Department of Defense Directive 3000.09 and the NATO Principles of Responsible Use of Artificial Intelligence in Defence (2021).
Accountability. The Customer shall name the Accountable Role in the Order Form and shall keep records of how the Service informed its decisions.
Training of users. The Customer shall train the people who use Outputs on what the probabilities mean and where the Model is weak, using the Documentation and the evaluation results Bynn provides.
Traceability. Every Output can be traced to its request, the Model version and the time, so that each decision can be reviewed afterwards.
Sessions and automated loops. coming soonWhere the Customer uses a Session, or any other automated loop, under this Addendum, including in connection with an unmanned aircraft, a vehicle or another system:
- (a) Outputs remain decision support under Section 3.5, and Section 7.6 to Section 7.8 of the Terms of Service apply;
- (b) no Output shall aim, trigger or release a weapon, or select or engage a target, without the authorisation of a responsible person under Section 5.1 for that use of force, and Section 4.8 applies to every Frame; and
- (c) the records under Section 5.2 and the traceability under Section 5.4 cover each Frame and its Output.
Section 6. The Safety Check, logging and audit
#The switch. Bynn may switch off the Harm and Targeting Component on the Named Keys. Only Bynn sets the switch; the Customer cannot. The Self-Harm Component always stays on.
Logging. For every request served with the Harm and Targeting Component switched off, Bynn logs the API Key, the time, a cryptographic hash of the request and the probability returned by the Safety Check. Bynn keeps these logs for two (2) years, and shall not access the content of a request except to investigate a suspected breach of this Addendum or as the law requires.
Audit. Bynn may audit the Customer's compliance with this Addendum on reasonable notice, within the Customer's security arrangements. The Customer shall give reasonable assistance, including access to its records under Section 5.2.
Section 7. Testing, reliability and bias
#Before use. Before operational use, the Customer shall test the Service for its intended use, including its accuracy, its calibration and its fairness across groups. Bynn provides its evaluation and bias results for the Model versions in use.
Pinned Versions. The Customer shall use a Pinned Version. Bynn shall give at least twelve (12) months' notice before retiring it, and shall publish evaluation results for each new version before the Customer switches to it.
Known limits. Bynn documents the Model's known weaknesses and updates that list when it finds new ones. The Customer shall take them into account in its procedures.
Section 8. Deployment, security and data
#Deployment. The Service is offered on Bynn's hosted API, on an instance dedicated to the Customer, or on the Customer's premises, including on networks with no connection to the internet. Where the Customer runs the Model on its premises:
- (a) this Addendum applies in full, and its terms form part of the licence under which Bynn provides the Model;
- (b) the Customer shall run the Safety Check software that Bynn supplies, with the Self-Harm Component switched on at all times, and shall not modify, disable or circumvent it;
- (c) the Customer shall itself keep the records required by Section 5.2 and Section 6.2 for two (2) years; and
- (d) the Customer shall give Bynn access to those records, and to the deployment, in an audit under Section 6.3.
Classified information. The Customer shall not send classified information to the hosted API. Classified use requires a deployment accredited for the relevant level, agreed in writing.
Customer Data. Bynn uses the Customer's Input and Output only to provide the Service and to perform this Addendum, and never uses them to train its Models.
Security. Each party shall protect the Service and the data with measures appropriate to the deployment, and shall notify the other without undue delay of any security incident that affects it.
Section 9. Incidents, suspension and termination
#Reporting. The Customer shall notify Bynn within seventy-two (72) hours to be confirmed of becoming aware of any use of the Service that may breach Section 4, and of any incident in which Outputs contributed to harm to Protected Persons.
Suspension. If Bynn reasonably believes that Section 4 has been breached, or that a deployment on the Customer's premises is not run as Section 8.1 requires, Bynn may switch the Harm and Targeting Component back on, suspend the Named Keys, suspend the licence of that deployment, or suspend this Addendum, in each case immediately, and shall tell the Customer why. Bynn has no liability for a suspension made in good faith under this Section 9.2.
Termination. Either party may terminate this Addendum for a material breach not cured within thirty (30) days of notice. Bynn may terminate it immediately for a breach of Section 4 or of export control law. On termination, all Military Use shall stop, and the Customer shall delete every copy of a Model held on its premises and certify the deletion in writing.
Survival. Section 4, Section 6.2, Section 6.3, Section 8.1(c), Section 8.1(d), Section 8.3, Section 9 and Section 10 survive the end of this Addendum.
Section 10. Responsibility, liability and law
#The Customer's responsibility. The Customer is solely responsible for every operational decision it makes with the Service, for every use of force, and for compliance with international humanitarian law and its rules of engagement.
Important: Disclaimer of warranties
10.2No warranty of fitness for military purposes. Bynn makes no warranty that the Service is fit for any military purpose. Section 7, Section 13, Section 14 and Section 15 of the Terms of Service apply to all use under this Addendum.
Indemnity. To the extent permitted by the law that applies to the Customer, the Customer's indemnity under Section 14.1 of the Terms of Service extends to every claim arising out of its Military Use of the Service.
Governing law and procurement rules. Where the Customer requires it, this Addendum, and the Agreement as it applies to the Customer, are governed by the law of the Customer's own state and are subject to the procurement rules that bind the Customer. Section 4, Section 6.3 and Section 9.2 form part of every contract under which the Service is provided for Military Use, apply whatever law governs that contract, cannot be waived or varied, and prevail over any conflicting term of any procurement contract, framework agreement, purchase order or standard terms.
Important: Limitation of liability
10.6Limitation of liability. Section 15.1 of the Terms of Service applies to this Addendum. Each party's total aggregate liability arising out of or in connection with this Addendum shall not exceed the Fees paid or payable by the Customer in the twelve (12) months before the event giving rise to the claim. This cap does not apply to liability for fraud, gross negligence or wilful misconduct, or breach of confidentiality; to Bynn's obligations under Section 14.2 of the Terms of Service; to the Customer's obligations under Section 10.3 and Section 14.1 of the Terms of Service, or its breach of Section 4; or to any liability that cannot be limited under the applicable law. This Section 10.6 replaces Section 15.2 of the Terms of Service only for liability arising out of or in connection with Military Use under this Addendum; for every other matter, Section 15.2 of the Terms of Service applies unchanged.
Insurance. Bynn shall maintain, with reputable insurers, insurance covering its liability under this Addendum with limits not less than the cap in Section 10.6, and shall provide evidence of that insurance on request.
Export classification. Before it signs its first Addendum, Bynn shall have the Models classified under the export control rules of the European Union and the United States by export counsel, with a formal ruling from the competent authorities where one is needed. No Model is provided under this Addendum before that classification is complete.
Classified work and clearances. Bynn shall seek the security clearances for its staff and facilities that classified work requires when the first Customer signs for classified use, with that Customer's government as sponsor. Until those clearances are granted, Bynn has no obligation to support classified use.
Points awaiting review
Open points for counsel and the owner, to be settled before this document takes effect. This list is removed when the document is published.
For counsel
- The export classification itself in the EU and the US (the commitment to classify before the first contract, Section 10.8, is decided).
- Indemnities specific to defence use, and how far a government Customer can give the indemnity in Section 10.3.
- The wording of the licence for deployments on the Customer's premises that carries this Addendum's terms (Section 8.1).
- Whether the EU Artificial Intelligence Act applies to any part of this use, given its exclusion for military purposes in Article 2(3).
- Section 5.5 (new): Sessions and other automated loops under this Addendum, including unmanned systems; the authorisation of a responsible person for every use of force a Frame informs, and records and traceability for each Frame.
- Whether Bynn's intellectual property indemnity and its liability for breach of confidentiality stay outside the twelve-month cap in Section 10.6 for defence Customers, against Section 14.4 of the Terms of Service, or take a separate higher cap, and the insurance limits in Section 10.7 that follow.
For the owner
- How soon the Customer must report a possible breach of Section 4 (72 hours drafted).
Values still open in this document
- The date on which these documents take effect
- How soon the Customer reports a possible breach of Section 4 of the Defence Contract Addendum (72 hours proposed)