Inscendo: AI Code of Conduct & AI Risk Disclosure

Effective Date: June 26, 2026 Version: 1.1 Provider: Inscendo Automation Inc.

This AI Code of Conduct ("Code") is incorporated into the Inscendo Master Subscription Agreement ("MSA") and applies to every Customer, user, and End-User of the Service. It governs (a) the AI risk disclosures Customer is on notice of and accepts, and (b) the obligations Customer assumes when deploying Inscendo IQ and any other AI feature of the Service. Violation of this Code is a material breach of the MSA. Capitalized terms have the meanings given in the MSA and the Acceptable Use Policy ("AUP").

---

1. AI Risk Disclosures Customer Acknowledges

Customer acknowledges and accepts each of the following:

1.1 Third-party model

The AI features of the Service are powered by foundation language models supplied by a third-party provider and may, in Inscendo's sole discretion, be powered by alternative or additional third-party model providers from time to time. Inscendo does not develop, train, or operate these foundation models. The behavior, accuracy, safety, content policies, and availability of these models are determined by their providers. Inscendo's commitments under the MSA are not, and shall not be construed as, warranties of the underlying model.

1.2 AI Output may be wrong

AI Output may be inaccurate, incomplete, outdated, biased, fabricated ("hallucinated"), discriminatory, infringing, defamatory, or otherwise harmful. AI Output that appears confident, well-formatted, or authoritative is not for that reason reliable. AI Output may invent citations, statutes, case law, financial figures, code APIs, medical guidance, person names, and historical facts. Customer is solely responsible for verifying AI Output before relying on, deploying, transmitting, or acting on it.

1.3 Inscendo IQ acts autonomously

Inscendo IQ is designed to (a) decompose Customer's instructions into multi-step plans, (b) call tools, including database mutations, schema changes, deletions, file writes, outbound HTTP calls, payment-processor calls, messaging providers, the Browser Companion, and any third-party API Customer has configured, and (c) execute those tools without contemporaneous human approval for each individual call, except for the limited, non-guaranteed confirmation behaviors described in § 2.1(a). Each agent action is an action of the Customer. Customer accepts that the agent may take actions Customer did not specifically anticipate but that are within the scope of Customer's instructions, configurations, or installed capsules.

1.4 Prompt injection and adversarial input

Inscendo IQ receives untrusted input from many sources, including end-user chat messages, web pages the Browser Companion observes, files Customer uploads, and responses returned by third-party APIs. Adversarial inputs, including prompts hidden in web-page text, document metadata, image alt-text, or API responses, can cause the agent to take actions different from those Customer intended, including actions that conflict with Customer's own instructions ("prompt injection"). Inscendo does not detect or block prompt-injection attempts and does not warrant that any will be prevented. Inscendo applies general least-privilege and containment controls, including server-side sandboxing of file and shell operations, allow-listing of outbound network destinations, gating of certain destructive database operations, and automated redaction of recognizable secrets from logs, that are designed to limit the impact of a successful injection; these controls reduce, but do not eliminate, risk, and no system is immune to every attack. Detection or filtering of harmful content, to the extent it occurs at all, is a function of the underlying third-party model and its provider, not of Inscendo. Because the Service does not prevent prompt injection, Customer is responsible for limiting the blast radius of a successful injection by controlling what the agent is exposed to, including the data, credentials, accounts, integrations, autonomy, and instructions Customer makes available to it, and by exercising the human oversight required by § 2.

1.5 Browser Companion-specific risks

The Browser Companion attaches to Customer's actual Chrome browser, with full read access to the active tab's contents, screenshots, console logs, network traffic, local storage, and the ability to evaluate JavaScript. The "sensitive-field sensor" automatically detaches the agent when standard password, one-time-code, and payment-card fields are detected; this is a heuristic and may fail to detect custom, non-standard, or programmatically rendered fields. Detached state may be reattached by Customer at any time. Data captured from the browser is transmitted to Inscendo's servers and to the AI model provider for processing. Because Inscendo IQ acts on what it observes on the page, it may be deceived by malicious or deceptive on-screen content, including hidden instructions; Customer should not direct the Companion at, or leave it attached to sessions logged into, financial, banking, healthcare, legal, government, or other sensitive accounts.

1.6 Logs and abuse monitoring

Inscendo retains operational and diagnostic telemetry for up to thirty (30) days for security, debugging, and abuse-detection purposes. Separately, Customer's conversation history and agent session records, including prompts, AI Output, tool calls and results, and browser observations, are retained as Customer Data for as long as the account is active and are deleted in accordance with MSA §§ 4.7–4.8. Inscendo applies automated secret redaction to telemetry. Inscendo does not apply automated PII redaction. Customer should not place sensitive personal information in inputs the agent will process unless Customer accepts that such information will be transmitted to the AI provider and retained as described.

1.7 No fitness for any particular purpose

The Service and AI Output are provided AS-IS. The Service has not been validated for any particular regulated use. Customer shall not represent to any End-User that AI Output is professional advice, regulated under any specific framework, or fit for any particular purpose.

1.8 Outage propagation

Where the underlying model provider experiences outage, latency, or content-policy changes, the Service may stop functioning, return errors, or behave inconsistently. Customer's automations and dependencies on AI features are subject to those upstream conditions.

1.9 AI Output may not be unique; generated code and configurations

Because of how generative AI works, AI Output is not unique to Customer, and other users may receive the same or substantially similar Output in response to the same or similar inputs. Inscendo claims, and grants Customer, no exclusivity in AI Output beyond the ownership allocation in MSA § 4. AI Output, including generated code, queries, schemas, capsules, widgets, automations, and configurations, may resemble, or be identical to, pre-existing third-party or open-source material, including material distributed under licenses that impose conditions (such as attribution or copyleft obligations). Customer is solely responsible for reviewing AI Output for third-party rights and license obligations, and for satisfying those obligations, before retaining, deploying, distributing, or relying on it.

2. Customer's Obligations Regarding AI Use

By using the AI features of the Service, Customer agrees to all of the following.

2.1 Human oversight

Customer shall implement human oversight commensurate with the risk and reversibility of the agent's actions. Customer shall:

(a) supervise the agent's actions with care proportionate to their risk and reversibility. Customer acknowledges that the Service executes the agent's tool calls, including destructive operations such as data mutations, schema changes, deletions, automation invocations that send outbound communications or move money, calls to external payment, messaging, or banking providers, and browser navigation, without human-verified confirmation of each individual call, and that the Service does NOT provide a general, customer-configurable system of confirmation gates or per-tool permissions. Customer must not assume such a system exists. The only confirmation behaviors the Service provides are: (i) Inscendo IQ, following its operating instructions, may prompt Customer to confirm before certain actions, but this is model-mediated and is not guaranteed; (ii) for certain database operations, destructive schema changes (e.g., dropping or truncating tables) and data mutations that would cascade into other automations, the Service interposes a confirmation step that is surfaced before the operation proceeds; and (iii) for background ("headless") agents Customer creates, Customer may constrain permitted writes through a mutation policy, and writes outside that policy are refused. Apart from these limited mechanisms, the means by which Customer limits the agent are operational, not configurable in the product: namely how Customer instructs, scopes, tests, and supervises the agent, and what data, credentials, accounts, and integrations Customer exposes to it. Accordingly, Customer shall not delegate irreversible or high-impact operations to the agent without Customer's own out-of-band controls, review, and ability to halt and roll back;

(b) review agent logs at a frequency proportionate to risk;

(c) test new automations and capsules in a non-production environment before exposing them to End-Users or production data;

(d) maintain an internal procedure for halting and rolling back agent actions in the event of malfunction;

(e) maintain identifiable Customer staff responsible for reviewing AI Output and agent decisions.

2.2 No consequential decisions without human review

Customer shall not delegate to Inscendo IQ any consequential decision, including without limitation decisions affecting an individual's:

• legal status, criminal-justice exposure, or immigration status;
• financial position (lending, credit limits, insurance underwriting, debt collection actions);
• employment opportunities (hiring, firing, promotion, demotion, compensation, scheduling, performance ratings);
• housing (tenant screening, rent decisions, eviction, mortgage approval);
• education (admission, financial aid, disciplinary action);
• healthcare (diagnosis, treatment, triage, prior authorization, pharmacy substitution);
• access to essential services (utilities, public benefits, banking);
• access to fundamental rights (voting, expression, assembly);

without meaningful human review of each individual decision by a competent human reviewer empowered to override the agent's recommendation. "Meaningful human review" means more than rubber-stamping; the reviewer must have the information, authority, and incentives to override the agent.

2.3 AI disclosure to End-Users

Where Customer's End-Users interact with the Service in a manner that (a) involves Inscendo IQ acting on the End-User's behalf or in response to the End-User's input, or (b) presents AI Output to the End-User as if it were content authored by Customer or a human, Customer shall clearly and conspicuously disclose to those End-Users that they are interacting with, or receiving content from, an AI system. This includes:

(i) a clear in-product label at the start of any agent-mediated chat or interaction; (ii) a persistent indicator throughout the interaction; (iii) where AI Output is delivered via email, SMS, voice, or other one-way channel, an in-content disclosure (e.g., "This message was generated by an AI assistant"); (iv) compliance with the Utah AI Policy Act safe-harbor disclosure requirements, the California AI Transparency Act (where Customer's user volume exceeds the statutory threshold), the Colorado AI Act, the Texas Responsible AI Governance Act, and any analogous state or federal law that comes into force.

2.4 No deceptive impersonation

Customer shall not configure Inscendo IQ to impersonate any specific named human, public figure, or entity without that person's verifiable consent. Customer shall not configure the agent to deny, when directly and sincerely asked, that it is an AI system.

2.5 No deepfakes; no synthetic identity fraud

Customer shall not use the Service to generate audio, video, or imagery depicting an identifiable real person without that person's verifiable consent and clear AI-disclosure labeling. Customer shall not use the Service to generate synthetic identities, fake government documents, or any artifact intended to commit fraud.

2.6 Input-rights warranty

Customer represents and warrants that Customer has all rights necessary to provide each input to the Service (including text, files, web-page contents Customer instructs the Browser Companion to capture, and third-party API responses Customer authorizes Inscendo IQ to ingest). Customer shall not provide as input any material Customer is not authorized to disclose, including third-party trade secrets, attorney-client-privileged communications, or third-party personal information lacking a lawful basis for processing.

2.7 Output-review duty

Before retaining, deploying, transmitting to any third party, publishing, selling, or causing Inscendo IQ to act on AI Output, Customer shall (a) review the AI Output for accuracy, bias, infringement, and harmful content; (b) verify any factual claim, citation, identifier, or computation; and (c) where AI Output includes generated code, queries, or configurations that will execute against Customer or End-User systems, review, test, and validate the code for correctness, security, and the absence of vulnerabilities before it runs. Customer acknowledges that AI Output, including generated code, may contain errors, security vulnerabilities, or infringing material. Any reliance on, execution of, or action taken on AI Output is at Customer's sole risk. Failure to perform this review is Customer's risk and is grounds for indemnification under MSA § 12.2.

2.8 Safety systems

Customer shall not (a) disable, weaken, modify, or attempt to circumvent any safety, security, or technical protection measure in the Service, including the Browser Companion sensitive-field sensor, rate limits, the database mutation gate, outbound-destination allow-listing, sandbox isolation, automated secret redaction, any confirmation prompt Inscendo IQ presents, or any content-moderation or safety control applied by the underlying third-party model provider, or (b) configure or instruct the agent to do so on Customer's behalf. Customer acknowledges that content moderation of model behavior, where it occurs, is performed by the third-party model provider and not by Inscendo.

2.9 Red-teaming and security research

Customer shall not perform adversarial testing, jailbreaking, prompt-injection research, or model-extraction attempts against the Service except (a) on Customer's own tenant, against Customer's own configurations, by Customer's own authorized personnel, and (b) consistent with the AUP. Customer shall report any vulnerability discovered in the Service to support@inscendoiq.com and shall not publicly disclose the vulnerability without coordination, except as required by law.

2.10 Bias, accessibility, and impact

Where Customer deploys an AI feature in employment, lending, housing, insurance, healthcare, education, or any other context covered by anti-discrimination law, Customer shall implement and document its own bias and impact testing, accessibility accommodations (including for users covered by the Americans with Disabilities Act), and any algorithmic-impact-assessment requirements applicable in Customer's jurisdiction.

2.11 Recordkeeping

Customer shall maintain records sufficient to demonstrate compliance with this Code, including (a) consent records for End-User communications, (b) AI-disclosure implementation, (c) human-review logs for consequential decisions, (d) bias and impact testing where applicable, and (e) incident-response records. Customer shall make such records available to Inscendo on reasonable request in the event of an investigation or claim.

2.12 No training of third-party models on Customer's behalf

Customer shall not use the Service to scrape, extract, or process content for the purpose of training a third-party AI model that competes with Inscendo's models or any provider whose terms of service Customer is bound by.

2.13 Least privilege and containment

Because the Service does not prevent prompt injection or guarantee that Inscendo IQ will act only as Customer intends, Customer shall apply the principle of least privilege so that any successful injection, error, or unintended action causes minimal damage. Customer shall: (a) grant the agent access only to the data, credentials, accounts, integrations, and autonomy strictly necessary for Customer's task, and withhold access the agent does not need; (b) not expose the agent to production data, money-movement, outbound communications, or sensitive systems without Customer's own oversight and the ability to halt and roll back; (c) require Customer's own human confirmation, out of band from the Service, before the agent takes any consequential or irreversible action (including payments, purchases, publishing, sharing personal data, or accepting terms on Customer's behalf); and (d) not rely solely on any safeguard Inscendo provides, Customer remains responsible for its own controls.

2.14 Autonomy does not reduce responsibility; provider requirements flow down

Operating Inscendo IQ in an autonomous, scheduled, or "headless" mode does not reduce or alter any of Customer's obligations under this Code, the AUP, or the MSA; Customer remains fully responsible for every action the agent takes on Customer's behalf, whether or not Customer specifically anticipated it. Customer's use of the Service must also comply with the acceptable-use, safety, and high-risk requirements of the third-party model provider(s) that power the Service, as Inscendo communicates or makes them available from time to time. Inscendo may impose, by notice, additional restrictions necessary to comply with its upstream providers' terms, and Customer shall comply with them.

3. Restricted Uses Requiring Inscendo Approval

The following uses are not flatly prohibited by the AUP but require Customer's written notice to Inscendo and Inscendo's written approval (which may be conditioned on additional terms or fees):

(a) high-volume content generation for public distribution (more than 100,000 unique pieces of public-facing content per month); (b) AI agents acting on behalf of an organization other than the Customer (e.g., where Customer operates the Service as an "agent-as-a-service" for client agencies); (c) generative AI features that create images, audio, or video in volumes that may trigger SB-942 (California AI Transparency Act) thresholds; (d) integration with election-related, voter-engagement, or political-advocacy operations; (e) deployment of Inscendo IQ in any context covered by the EU AI Act high-risk classifications, even if Customer's End-Users are U.S.-based but the deployment may be observed or used in the EU; (f) deployment of the agent in critical infrastructure adjacencies (energy load forecasting, water-utility billing, transit-scheduling).

4. Updates

Inscendo may update this Code from time to time as the AI legal and regulatory landscape evolves. Material adverse changes will be notified per MSA § 15.

[End of AI Code of Conduct]