AI/ML
Cloud AI and Model-as-a-Service (MaaS) Service Terms
PART I – SERVICE DESCRIPTION AND LEGAL BASIS
▸ Legal basis:
- Cybersecurity Law 2018 (Articles 8, 16);
- The Penal Code 2015 (Bộ luật Hình sự 2015) (Articles 174, 288, 316);
- Decree 13/2023/NĐ-CP (Articles 11 and 17);
- EU AI Act 2024 (Article 5 – prohibited AI systems);
Article 1. Definitions and scope of AI services
1.1. Cloud AI and MaaS (Model-as-a-Service) provide API access to artificial intelligence models developed by HiTechCloud or integrated from leading AI providers, including but not limited to:
- (a) Natural language processing models (NLP/LLM): text generation, translation, summarization, sentiment analysis;
- (b) Computer vision models: image recognition, OCR and video analysis;
- (c) Speech AI models: speech recognition (STT) and speech synthesis (TTS);
- (d) Generative AI: creating images, video, audio and source code;
- (e) Analytical AI: forecasting, anomaly detection and data clustering.
1.2. HiTechCloud acts as an intermediary API provider (API Aggregator/Reseller) for third-party AI models, and as the direct developer of its own in-house AI models.
▸ Legal basis: Decree 13/2023/NĐ-CP (personal data processing in AI); the Cybersecurity Law 2018 (Luật An ninh mạng 2018, Article 16 on AI content); the National AI Strategy to 2030 (Decision 127/QĐ-TTg 2021); the EU AI Act 2024 (applicable to EU users and models trained on EU data); Decree 147/2024/NĐ-CP.
Article 2. Cloud AI / MaaS service plans
| Plan | API calls/month | Tokens/month | Model | Latency SLA | Price |
| AI Starter | 100.000 calls | 5M tokens | Basic model | < 2s P95 | Per the price list |
| AI Business | 1M calls | 50M tokens | Basic + advanced | < 1s P95 | Per the price list |
| AI Professional | 10M calls | 500M tokens | Full catalog | < 500 ms P95 | Per the price list |
| AI Enterprise | Unlimited | As agreed | Fine-tuned model | Custom SLA | Get a quote |
| Pay-as-you-go | Based on actual usage | Based on actual usage | Theo model | Best effort | $/1K tokens |
PART II – CONDITIONS OF AI USE
Article 3. Permitted use cases
3.1. The Customer may use Cloud AI/MaaS for lawful purposes, including:
- (a) Business applications: customer support chatbots and robotic process automation (RPA);
- (b) Creative content: producing marketing copy, translation and text summarization;
- (c) Data analysis: customer sentiment analysis and demand forecasting;
- (d) Education applications: personalized learning systems and automated assessment;
- (e) Lawful scientific research and technology development;
- (f) Clinical decision support systems (not a replacement for a doctor's medical judgment).
Article 4. Strictly prohibited uses
4.1. It is strictly prohibited to use Cloud AI/MaaS to:
- (a) Generating child sexual abuse material (CSAM) or exploiting or abusing children in any form;
- (b) Creating deepfakes of real people without consent, particularly for unlawful purposes;
- (c) Creating organized disinformation intended to cause social harm;
- (d) Circumventing or defeating security systems, CAPTCHA or authentication;
- (e) Creating malware, attack code or security vulnerability exploits;
- (f) Profiling individuals or tracking them without consent;
- (g) Discriminatory social credit scoring systems;
- (h) Autonomous weapons and systems that make decisions harmful to people without human oversight;
- (i) Financial fraud, phishing and AI-automated social engineering;
- (j) Impersonating a real person's voice or likeness for fraud or manipulation;
- (k) Processing an individual's biometric data without explicit consent.
Article 5. Responsibility for AI inputs and outputs
5.1. Responsibility for input/prompt:
- (a) The Customer bears full legal responsibility for prompt content and input data;
- (b) Sensitive personal data of third parties must not be ingested without a lawful basis;
- (c) Copyrighted data must not be ingested for retraining or exploitation without permission.
5.2. Responsibility for output/response:
- (a) AI output is probabilistic and may be inaccurate. The Customer is responsible for reviewing and verifying it before relying on it for material decisions;
- (b) HiTechCloud does NOT warrant that AI output is accurate, complete or fit for purpose;
- (c) AI output does not constitute professional advice (medical, legal or financial) and must not be used as a substitute for expert advice;
- (d) The Customer owns the output AI generates from their input (except where it infringes a third party's IP rights);
- (e) HiTechCloud is not liable for actions taken or damage arising from the use of AI output.
Article 6. Training data and AI privacy
6.1. Data policy for AI services:
- (a) By default, API input and output data is NOT used to retrain AI models;
- (b) API data is held in cache for a maximum of 30 days and then deleted;
- (c) Customers can enable “Zero Data Retention” (ZDR), under which no data passing through the API is stored (Business+ plans);
- (d) On the Enterprise plan, models can be deployed in a private environment, with no data leaving the Customer's infrastructure.
6.2. Fine-tuning and Custom Models:
- (a) Where the Customer supplies data to fine-tune a model, that data remains the property of the Customer;
- (b) The fine-tuned model is the Customer's exclusive property and is not shared with other customers;
- (c) HiTechCloud uses fine-tuning data for the sole purpose of training that specific Customer's model.
Article 7. API usage, quotas and billing
7.1. AI service fees are based on:
- (a) Number of API calls;
- (b) Tokens (units of text processing): input tokens + output tokens;
- (c) Compute time: applies to compute-intensive tasks (image generation, video);
- (d) AI storage: storing fine-tuned models and vector databases.
7.2. Rate limiting:
| Plan | RPM (requests/minute) | TPM (tokens/minute) | Concurrent connections |
| AI Starter | 60 RPM | 60.000 TPM | 5 |
| AI Business | 600 RPM | 600.000 TPM | 20 |
| AI Professional | 6.000 RPM | 6.000.000 TPM | 100 |
| AI Enterprise | Customization | Customization | Unlimited |
7.3. When the rate limit is reached, the API returns HTTP 429 (Too Many Requests). The Customer must implement retry logic with exponential backoff in the application.
Revision history
Current version
Updated
Updated
Updated
Updated
Category: AI/MLGet notified when new documents are added to this category.
