AI Output and Safety

1. Purpose

VOSCOM-ONLINE LLC (ООО «ВОСКОМ-ОНЛАЙН»), Russian Federation; OGRN 1217700640189, INN 9719021912, KPP 771901001. Registered address: 105484, Moscow, municipal district Vostochnoye Izmaylovo, Sirenevy Boulevard, building 50, apartment 72.

This notice supplements the Terms, Licence and AUP. Requirements for human review, laboratory testing and not relying on AI as the sole assurance of safety are retained.

2. What Constitutes AI Output

2.1. For purposes of VOSCOM.ONLINE, “AI Output” means any results generated or suggested, in whole or in part, using AI features, orchestration logic, data search, rules, manifests, templates, compatibility logic, or other automated mechanisms of the Platform.

2.2. AI Output may include:

• design prompts, explanations, compatibility hints, and recommendations;
• DSL/JSON/YAML manifests, project manifests, configuration structures, pin maps, module graphs, and dependency suggestions;
• code suggestions, build scripts, firmware artifacts, sample implementations, code fragments, device settings, and connection diagrams;
• results of cloud builds, compatibility checks, static checks, reports, comparisons, warnings, and other automated assessments;
• outputs prepared in the chat interface, API, playground, user account, monitoring module, documentation assistant, or other Platform interfaces.

2.3. The existence of AI Output does not mean that it is a verified engineering solution, an approved project, a certified architecture, a safe configuration, or an artifact suitable for immediate deployment.

3. Instrumental, Not Certifying, Nature of the Platform

3.1. VOSCOM.ONLINE is an engineering assistance tool and not an automated substitute for a designer, developer, tester, verifier, safety engineer, certification body, conformity assessment body, or independent laboratory validation.

3.2. Any AI Output is provided as a working hypothesis, intermediate artifact, routine-operation accelerator, or convenient interface to the Platform’s libraries, rules, and service logic.

3.3. The user may not interpret AI Output as:

• a legally binding warranty of operability;
• a certificate of safety, compatibility, or compliance with an industry standard;
• confirmation that a device, firmware, or configuration meets the requirements of a customer, regulator, contract, insurer, manufacturer, EPC contractor, or system integrator;
• authorization to use the solution in industrial operation, field deployment, or a safety-critical environment without the user’s own review and approval cycle.

4. AI Limitations and Possible Causes of Errors

4.1. Even where curated modules, manifests, compatibility rules, validation layers, and other safeguards are used, AI Output may contain errors, omissions, unsafe assumptions, incorrect dependencies, incomplete edge cases, conflicting settings, false compatibility conclusions, or other defects.

4.2. In particular, errors may arise due to:

• incomplete, ambiguous, or incorrect user input;
• incomplete information about equipment, board revision, datasheet, pinout, power budget, signal integrity, electromagnetic compatibility, temperature conditions, operating environment, mounting mechanics, or facility constraints;
• discrepancies between the actual device and the model, template, module, or description in the Platform;
• errors in third-party components, external libraries, build tools, SDKs, API providers, model providers, or user modifications;
• automated reasoning errors, LLM variability, hallucinations or similar behavior, non-determinism, gaps in data retrieval, outdated documentation, or incorrect assumptions;
• the fact that certain critical parameters were not provided to the Platform at all and therefore could not be taken into account by the system.

4.3. For these reasons, AI Output must never be considered a sufficient basis for direct deployment without human review and subsequent engineering verification.

5. Mandatory Human Review

5.1. Before any practical use of any AI Output, you must ensure review by a competent human who has sufficient knowledge and authority to assess the specific use case.

5.2. Such review must include, to the extent applicable:

• verification of the solution logic and its conformity with the task requirements;
• review of the connection diagram, pinout, voltage levels, power supply, current limits, timings, input/output modes, interfaces, and module compatibility;
• review of the initial assumptions, operating conditions, environmental constraints, fail-safe logic, watchdog behavior, restart behavior, OTA behavior, and recovery paths;
• review for compliance with internal engineering procedures, project documentation, customer requirements, contract terms, industry standards, quality assurance rules, and change management procedures;
• confirmation that the AI Output is an acceptable starting point for subsequent testing and not a final result ready for deployment.

5.3. If you do not have such competence, or if the project involves elevated risk, the AI Output must not be used until it has undergone independent qualified review.

6. Mandatory Testing and Validation Before Deployment

6.1. Before flashing, launch, delivery, handover to a customer, commercial deployment, field deployment, or any other practical use, you must conduct your own testing and validation.

6.2. Unless your process provides otherwise, the minimum expected scope of checks includes:

• build verification and control of build reproducibility;
• unit and integration testing, where applicable to the specific project;
• bench testing on real hardware or an equivalent test bench;
• power and thermal checks, restart tests, communication tests, timeout tests, fault injection tests, and degraded-mode tests;
• checking for configuration errors, swapped pins, incorrect signal levels, overflows, race conditions, deadlocks, resource contention, module incompatibility, and incorrect behavior under fault conditions;
• validation at the site or in an environment as close as possible to real operation before industrial launch.

6.3. For regulated, industrial, distributed, hazardous, gas-related, utility, monitoring, security, infrastructure, and other serious use cases, the list of mandatory tests must be substantially broader and must be determined by your own engineering responsibility framework.

7. What the Platform Does Not Guarantee

7.1. The Platform does not guarantee that AI Output:

• is fully correct, safe, optimal, or free from defects;
• is suitable for industrial operation, serial production, or critical operation;
• takes into account all actual properties of the equipment, environment, facility, and context;
• meets the requirements of a specific customer, industry standard, regulator, insurer, or audit;
• does not infringe third-party rights, open-source license obligations, contractual restrictions, or export regimes.

7.2. No successful builds, compatibility labels, absence of warnings, low-risk indicators, proposed topologies, module datasheets, community feedback, sample projects, or marketing materials shall be interpreted as a universal or absolute warranty of operability.

8. No Certification Unless Separately and Expressly Issued

8.1. Unless you have been issued a separate written document confirming certification, conformity assessment, laboratory validation, compatibility confirmation, enterprise approval, or another formalized act, you may not claim that the Platform, its AI Output, or any specific generated artifact is certified, approved, suitable for regulated use, or officially authorized for deployment.

8.2. Any labels such as verified, compatible, passed, recommended, or similar designations must be interpreted strictly within the scope expressly described in the interface, documentation, or separate agreement, and may not be expanded by the user at their own discretion.

9. Prohibited and Restricted Use Cases

9.1. Without a separate written agreement expressly permitting the relevant scope of use, VOSCOM.ONLINE and AI Output are not intended to be used as the sole, final, or unconditionally trusted solution in the following scenarios:

• medical devices, clinical environments, diagnostics, treatment, life support, or other medically regulated systems;
• aviation, space systems, railway safety, autopilot systems, transportation safety, emergency response, firefighting, and other life-critical systems;
• nuclear infrastructure, weapons, combat systems, unlawful surveillance, offensive cyber operations, and other highly hazardous or prohibited end uses;
• industrial facilities where an error may cause an explosion, leak, environmental damage, injury, substantial property damage, or shutdown of critical production;
• systems requiring mandatory certification, formal methods, regulatory approval, accredited laboratory testing, or other special procedures, unless such procedures have been separately completed.

9.2. Even if a specific scenario is not listed above, it shall be deemed restricted if a reasonable engineer could conclude that an error in AI Output in that scenario creates a material risk to life, health, property, the environment, infrastructure, or third-party rights.

10. Responsibility of the User, Integrator, and Person Implementing the Result

10.1. The user, integrator, customer, seller, OEM, system integrator, engineer, or any other person implementing the result who actually uses AI Output is responsible for:

• the correctness of the task statement and completeness of the input data;
• reviewing all engineering assumptions and constraints;
• conducting a sufficient level of testing and validation;
• deciding whether the result may be deployed in a specific facility, product, line, device, or customer environment;
• complying with applicable laws, industry standards, contracts, licenses, safety rules, cybersecurity rules, export rules, and personal data requirements.

10.2. If you transfer AI Output to a third party, sell a solution, include the result in a commercial product, or use it in design or deployment for a customer, you are solely responsible for describing and communicating all applicable limitations, risks, human review requirements, and warranty limitations.

11. Third-party providers and BYOK

When connecting a personal API key or third-party AI service, users choose the provider and must consider its terms and data practices. Do not supply materials or third-party data without the required rights and legal basis. The platform sends the selected model the request and necessary context, including conversation history and requested project excerpts. This does not remove the Operator’s own data-protection duties; consent does not replace specific cross-border transfer requirements.

12. Monitoring, Warnings, and Automated Checks Do Not Release You from Responsibility

12.1. The Platform may use automated checks, static checks, compatibility checks, pin-conflict warnings, policy screens, anti-abuse filters, usage notices, build diagnostics, and other warning layers.

12.2. The existence of such checks does not mean that the system detects all errors and risks.

12.3. The absence of a warning, a successful build, passing a validation step, a positive status, a green indicator, or any other automated mark does not mean that the solution is safe or suitable for deployment.

13. Reporting Incidents and Unsafe Behavior

13.1. If you discover that any AI Output, compatibility logic, sample project, generated artifact, module datasheet, template, or other part of the Platform:

• creates a material safety risk;
• may lead to a dangerous configuration;
• misleads the user regarding safe use;
• incorrectly marks a solution as compatible, safe, or recommended;
• contains an obvious error capable of causing significant harm,

you must report it without undue delay to admin@voscom.online.

13.2. We may, at our discretion, restrict, disable, revoke, unpublish, correct, or change any AI Output, example, module, rule set, compatibility note, or feature if we deem it necessary to reduce risk.

14. Right to Restrict Unsafe Use

14.1. We may restrict, suspend, or terminate the use of AI features, APIs, build environments, modules, manifests, compatibility services, or the entire account if we believe that the user:

• uses the Platform in an unsafe, abusive, or prohibited manner;
• attempts to apply AI Output in prohibited or unauthorized critical scenarios;
• circumvents warnings, policy screens, safety prompts, validation logic, or other protective controls;
• ignores mandatory engineering review and distributes AI Output as allegedly guaranteed safe or certified.

15. Liability and dangerous-action confirmation

Risk warnings do not exclude statutory Operator liability. Check the processor model, write addresses, keys and consequences of eFUSE changes. Irreversible operations and device writes require separate explicit human confirmation, not merely consent to an AI conversation.

Mandatory consumer rights and safeguards under Russian law prevail. Nothing in this document waives rights that cannot be restricted by agreement, including remedies for defective services and refunds where required by law.

16. Changes

New editions are published with a version and effective date; earlier editions remain in the history. Material changes are notified in the interface. Separate consent is requested where necessary; using the site does not itself provide personal-data or public-disclosure consent. Changes do not impair accrued mandatory rights or alter an agreed paid price or term without a lawful basis.

17. Contacts

VOSCOM-ONLINE LLC (ООО «ВОСКОМ-ОНЛАЙН»), Russian Federation; OGRN 1217700640189, INN 9719021912, KPP 771901001. Registered address: 105484, Moscow, municipal district Vostochnoye Izmaylovo, Sirenevy Boulevard, building 50, apartment 72.

Support: support@voscom.online. Security incidents: admin@voscom.online.