DeepSeek
deepseek.com
Automated, point-in-time. Not independently reviewed or approved by TRUSTYCYBER unless expressly stated.
- Official-model inputs and outputs are used for training by default; opt-out is by email only
- No formal subprocessor register disclosed
- No clear commitment that your data will not train their models
+3 more
See Before you sign, with what to ask for ↓
Scanned Oct 21, 2026 · Public evidence · Point-in-time
Higher scores indicate stronger, independently supported AI assurance, governance, AI systems, models, data, supply chain and third-party assurance evidence. The score and the decision are related but different: a well-scored vendor can still carry conditions.
Before you sign
Why it matters: In Official Model mode the Chinese agreement states DeepSeek may use collected inputs and outputs for model training and service optimisation after de-identification, with the user granting a worldwide licence; the only opt-out is to contact the vendor. No retention period for ordinary prompts and outputs is given, and records tied to suspected violations are retained and may be reported to authorities. Processing location is not disclosed and the contract is governed by mainland PRC law.
What to ask for: Please confirm in writing: (a) whether Official Model mode inputs/outputs are used to train or fine-tune models by default, (b) the retention period for inputs and outputs after opt-out, (c) the country or region in which Harness inputs are processed and stored, and (d) whether an organisation-level training opt-out can be applied contractually rather than per user by email.
“（3）为向您提供连续、高质量的服务， 在经安全加密技术处理、严格去标识化且无法重新识别特定个人的前提下，我们可能会将服务所收集的输入及对应输出，用于模型训练和服务的优化。在此前提下，您同意授予深度求索在全球范围内一项免费的、非排他的许可使用权，以使我们对前述内容进行处理。 如您拒绝将您的数据用于模型训练，可以通过第七条的联系方式联系我们来选择退出，关闭后您的输入和输出不会再用于我们的模型训练。”
“(3)Under the premise of secure encryption technology processing, strict de-identification rendering, and irreversibility to identify specific individuals, we may, to a minimal extent, use Inputs and Outputs to provide, maintain, operate, develop or improve the Services or the underlying technologies supporting the Services. If you refuse to allow us to process the data in the manner described above, you can opt out by contacting us through the contact information in Article VIII.”
“For behaviors suspected of violating laws and regulations or involving illegal activities, relevant records will be retained, and reports will be made to the competent authorities in accordance with the law, cooperating with their investigations.”
“1.The establishment, execution, interpretation, and resolution of disputes under these Terms shall be governed by the laws of the People's Republic of China in the mainland.”
Why it matters: The vendor names AI-related providers or partially addresses its supply chain, but the public sources scanned do not include a formal, dated subprocessor register.
What to ask for: Request a dated subprocessor register with change-notification terms for the DPA/order form.
Why it matters: The public sources scanned do not clearly state that customer prompts, files, and outputs are excluded from training or fine-tuning of first- or third-party models.
What to ask for: Get a no-training clause covering first- and third-party models into the DPA/order form.
“（3）为向您提供连续、高质量的服务， 在经安全加密技术处理、严格去标识化且无法重新识别特定个人的前提下，我们可能会将服务所收集的输入及对应输出，用于模型训练和服务的优化。在此前提下，您同意授予深度求索在全球范围内一项免费的、非排他的许可使用权，以使我们对前述内容进行处理。 如您拒绝将您的数据用于模型训练，可以通过第七条的联系方式联系我们来选择退出，关闭后您的输入和输出不会再用于我们的模型训练。”
“(3)Under the premise of secure encryption technology processing, strict de-identification rendering, and irreversibility to identify specific individuals, we may, to a minimal extent, use Inputs and Outputs to provide, maintain, operate, develop or improve the Services or the underlying technologies supporting the Services. If you refuse to allow us to process the data in the manner described above, you can opt out by contacting us through the contact information in Article VIII.”
Why it matters: The public sources scanned do not give a retention period for customer data or prompts/outputs.
What to ask for: Get retention windows, in days, in writing.
“3.Your Input will be sent directly to the corresponding AI model for processing, and the Output will also be returned directly by that AI model. DeepSeek Harness does not store or process your Input and Output on the server side.”
“For behaviors suspected of violating laws and regulations or involving illegal activities, relevant records will be retained, and reports will be made to the competent authorities in accordance with the law, cooperating with their investigations.”
Why it matters: No SOC 2 Type II, ISO/IEC 42001, or registry-verified certification covering the AI product was found in the public sources scanned.
What to ask for: Request a SOC 2 Type II or ISO/IEC 42001 report and confirm its scope covers the AI product.
Why it matters: The public sources scanned do not state where customer data is processed or offer a residency option.
What to ask for: Pin processing regions per data classification in the contract.
Buyer questionsanswered only from the public evidence scanned, click a row for the answer
!Will they train on your data?Ask the vendor
The Chinese-language agreement states that, after encryption and strict de-identification, DeepSeek may use collected inputs and corresponding outputs for model training and service optimisation, and the user grants a worldwide royalty-free licence for this; opt-out is only by contacting the vendor via the Section 7 contact details.
Requires written confirmation, see Before you sign ↓
“（3）为向您提供连续、高质量的服务， 在经安全加密技术处理、严格去标识化且无法重新识别特定个人的前提下，我们可能会将服务所收集的输入及对应输出，用于模型训练和服务的优化。在此前提下，您同意授予深度求索在全球范围内一项免费的、非排他的许可使用权，以使我们对前述内容进行处理。 如您拒绝将您的数据用于模型训练，可以通过第七条的联系方式联系我们来选择退出，关闭后您的输入和输出不会再用于我们的模型训练。”
“(3)Under the premise of secure encryption technology processing, strict de-identification rendering, and irreversibility to identify specific individuals, we may, to a minimal extent, use Inputs and Outputs to provide, maintain, operate, develop or improve the Services or the underlying technologies supporting the Services. If you refuse to allow us to process the data in the manner described above, you can opt out by contacting us through the contact information in Article VIII.”
!How long do they keep your data?Ask the vendor
In Custom Model mode inputs go directly to the user-configured model and outputs come straight back; DeepSeek Harness states it does not store or process Input and Output on the server side.
Requires written confirmation, see Before you sign ↓
“3.Your Input will be sent directly to the corresponding AI model for processing, and the Output will also be returned directly by that AI model. DeepSeek Harness does not store or process your Input and Output on the server side.”
“For behaviors suspected of violating laws and regulations or involving illegal activities, relevant records will be retained, and reports will be made to the competent authorities in accordance with the law, cooperating with their investigations.”
!Who else can access your data?Ask the vendor
In the default Official Models mode DeepSeek states it provides the full chain itself, from the foundation model to the interaction medium, using DeepSeek official models; no third-party model provider or hosting subprocessor is named for this mode.
Requires written confirmation, see Before you sign ↓
“Official models : When you use the DeepSeek official models configured by default after registering and logging in, we provide you with a full-chain service from the foundation model to the interaction medium.”
!Where is your data processed?Ask the vendor
Not found in the public sources scanned — ask the vendor directly.
!What happens in a security incident?Ask the vendor
DeepSeek asks that security vulnerabilities in Harness be reported privately to [email protected]; no customer breach-notification commitment or incident response process is described.
Requires written confirmation, see Before you sign ↓
“If you discover a security vulnerability, we recommend that you report it privately through the repository at [email protected]”
Email to send the vendor8 items to confirm in writing
Hello DeepSeek team, We are assessing DeepSeek as part of our supplier review. Before we proceed, please confirm the following in writing: 1. What is your commitment to notify customers of a security incident affecting our data, including the timeframe? 2. Please provide your current, dated subprocessor list and explain how you notify customers of changes. 3. Do you use our data (inputs, outputs or uploaded files) to train or improve your models? Please confirm in writing, including any opt-out and whether it applies to every plan. 4. How long do you retain our data, in days, for each type (inputs, outputs, logs and backups), and how is it deleted? 5. Please address the following: No independent assurance evidenced. 6. In which countries and regions is our data processed and stored, and can processing be limited to a region we choose? 7. In Custom Model mode, does any telemetry, usage data or prompt metadata flow to DeepSeek servers, and is the Harness client fully functional offline from DeepSeek infrastructure? 8. AWS appears to be involved as a platform provider: confirm whether this dependency exists, and whether it processes customer data. A written reply to each point, or a link to where it is documented, is enough. Where a point is covered by a certification or independent report, please include the certificate number or the report and its date. Thank you,
Key findingsclick a row for the evidence
✓Unusually candid disclosure of agentic security riskStrong
The Harness terms spell out that the agent executes model-generated code, loads plugins and touches credentials and files; that prompt injection remains an inherent risk; that sandboxes and approval prompts do not guarantee isolation; and that the service is experimental and not production-ready. They prescribe least-privilege, disposable environments, backups and human confirmation, and give a private vulnerability-reporting address.
Most agent-tool vendors do not state these limits in binding terms. A buyer can lift the vendor's own precaution list straight into an internal usage standard, and the 'not ready for production environments' statement should gate any deployment decision.
“The Services can access a variety of systems and can execute model-generated code and commands on behalf of the user, load third-party plugins, and access networks, processes, credentials, and files made available to them. Although most foundation models include basic defenses against prompt injection, inherent risks remain when using DeepSeek Harness to interact with the internet or to operate through other untrusted data sources.”
“Sandboxes, approval prompts, and permission controls can reduce risk, but they do not guarantee isolation and cannot guarantee that harm will be prevented. Even where restrictions are correctly enforced, they cannot protect resources that the project has been permitted to access.”
“Harness and related technologies such as agents are still at an early stage of development and are innovative and experimental in nature. The DeepSeek Harness Services are also still in a continuous testing phase and must not be regarded as fully secure or ready for production environments.”
“If you discover a security vulnerability, we recommend that you report it privately through the repository at [email protected]”
✗Official-model inputs and outputs are used for training by default; opt-out is by email onlyGap
In Official Model mode the Chinese agreement states DeepSeek may use collected inputs and outputs for model training and service optimisation after de-identification, with the user granting a worldwide licence; the only opt-out is to contact the vendor. No retention period for ordinary prompts and outputs is given, and records tied to suspected violations are retained and may be reported to authorities. Processing location is not disclosed and the contract is governed by mainland PRC law.
A coding agent routinely sees proprietary source code, credentials and internal data. Default-on training use with a manual opt-out, no stated retention window and no processing-location disclosure is a material data-governance gap for any organisation handling confidential or regulated information.
Question for vendor: Please confirm in writing: (a) whether Official Model mode inputs/outputs are used to train or fine-tune models by default, (b) the retention period for inputs and outputs after opt-out, (c) the country or region in which Harness inputs are processed and stored, and (d) whether an organisation-level training opt-out can be applied contractually rather than per user by email.
“（3）为向您提供连续、高质量的服务， 在经安全加密技术处理、严格去标识化且无法重新识别特定个人的前提下，我们可能会将服务所收集的输入及对应输出，用于模型训练和服务的优化。在此前提下，您同意授予深度求索在全球范围内一项免费的、非排他的许可使用权，以使我们对前述内容进行处理。 如您拒绝将您的数据用于模型训练，可以通过第七条的联系方式联系我们来选择退出，关闭后您的输入和输出不会再用于我们的模型训练。”
“(3)Under the premise of secure encryption technology processing, strict de-identification rendering, and irreversibility to identify specific individuals, we may, to a minimal extent, use Inputs and Outputs to provide, maintain, operate, develop or improve the Services or the underlying technologies supporting the Services. If you refuse to allow us to process the data in the manner described above, you can opt out by contacting us through the contact information in Article VIII.”
“For behaviors suspected of violating laws and regulations or involving illegal activities, relevant records will be retained, and reports will be made to the competent authorities in accordance with the law, cooperating with their investigations.”
“1.The establishment, execution, interpretation, and resolution of disputes under these Terms shall be governed by the laws of the People's Republic of China in the mainland.”
!Chinese and English terms differ on training use and input reviewGap
The Chinese agreement says inputs and outputs may be used for model training (用于模型训练) and that DeepSeek may review inputs and outputs for compliance. The English Terms of Use, dated the same day, soften this to 'to a minimal extent, use Inputs and Outputs to provide, maintain, operate, develop or improve the Services or the underlying technologies' and omit the explicit reference to reviewing inputs and outputs. The Chinese text carries explicit update and effective dates; the English shows only a last-update date.
An English-speaking buyer reading only the English terms would understate both the training use and the content inspection the vendor reserves. Where two language versions diverge, the buyer needs to know which governs; neither document states a precedence rule between languages.
Question for vendor: Which language version of the Harness terms prevails in the event of a conflict, and does the English 'develop or improve the Services or the underlying technologies' include training or fine-tuning DeepSeek models?
“（3）为向您提供连续、高质量的服务， 在经安全加密技术处理、严格去标识化且无法重新识别特定个人的前提下，我们可能会将服务所收集的输入及对应输出，用于模型训练和服务的优化。在此前提下，您同意授予深度求索在全球范围内一项免费的、非排他的许可使用权，以使我们对前述内容进行处理。 如您拒绝将您的数据用于模型训练，可以通过第七条的联系方式联系我们来选择退出，关闭后您的输入和输出不会再用于我们的模型训练。”
“(3)Under the premise of secure encryption technology processing, strict de-identification rendering, and irreversibility to identify specific individuals, we may, to a minimal extent, use Inputs and Outputs to provide, maintain, operate, develop or improve the Services or the underlying technologies supporting the Services. If you refuse to allow us to process the data in the manner described above, you can opt out by contacting us through the contact information in Article VIII.”
“（1）为履行法定义务的要求，深度求索有权采取技术手段等措施对用户使用本服务的行为及信息进行审查，包括但不限于对输入和输出进行审查、建立风险过滤机制、建立违法内容特征库等。作为本服务的使用者，”
“更新日期： 2026年9月20日 生效日期： 2026年9月27日 亲爱的用户，欢迎您使用DeepSeek Harness 产品及服务！ DeepSeek Harness 产品及服务由杭州深度求索人工智能基础技术研究有限公司（以下简称“我们”或“深度求索”）所有和负责运营。在使用本服务前，请您务必仔细阅读并理解本《DeepSeek”
!No independent assurance, named subprocessors or incident-notification commitmentGap
None of the three documents names a certification, audit report (SOC 2, ISO/IEC 27001, ISO/IEC 42001) or external evaluation, and the only security commitment is 'measures no less rigorous than industry practice'. Hosting and infrastructure subprocessors for Official Model mode are not named. The only incident-related mechanism is a vulnerability-reporting mailbox; there is no customer breach-notification undertaking. Independent assurance is assessed as not_evidenced rather than not_applicable because DeepSeek runs the full hosted chain in Official Model mode and could supply such evidence; subprocessors is assessed as partial rather than not_applicable for the same reason, even though Custom Model mode takes DeepSeek out of the request path.
Without third-party attestation or a subprocessor list, the buyer has only the vendor's own word for platform security and cannot map where data flows in hosted mode. Absence of a breach-notification clause means the buyer's own incident obligations (for example under APRA CPS 234 or GDPR Article 33) would not be supported contractually.
Question for vendor: Can DeepSeek provide a current SOC 2 Type II or ISO/IEC 27001 report covering the Harness hosted service, a list of hosting and infrastructure subprocessors, and a contractual security-incident notification commitment with a defined timeframe?
“We will take necessary measures no less rigorous than industry practice to ensure the cybersecurity and stable operation of the platform, and will strive to enhance and improve our technology to ensure your product and service experience.”
“Official models : When you use the DeepSeek official models configured by default after registering and logging in, we provide you with a full-chain service from the foundation model to the interaction medium.”
“If you discover a security vulnerability, we recommend that you report it privately through the repository at [email protected]”
✓Custom Model mode keeps inputs and outputs off DeepSeek serversStrong
When the user supplies their own model API, Harness states it does not store or process inputs or outputs server-side; data goes directly to the user's chosen provider. DeepSeek also publishes a Transparency Center listing model cards and technical reports for DeepSeek-V4 and DeepSeek-V3.2, and references a regulatory algorithm-filing disclosure.
Custom Model mode is a practical way to use the Harness tooling while keeping data within an already-assessed model provider, avoiding the Official Model mode training and retention concerns. The buyer then inherits the data terms of whichever provider they configure, and takes on responsibility for vetting third-party plugins, MCP servers, Skills and Hooks, which DeepSeek disclaims.
Question for vendor: In Custom Model mode, does any telemetry, usage data or prompt metadata flow to DeepSeek servers, and is the Harness client fully functional offline from DeepSeek infrastructure?
“3.Your Input will be sent directly to the corresponding AI model for processing, and the Output will also be returned directly by that AI model. DeepSeek Harness does not store or process your Input and Output on the server side.”
“Custom models : You may also configure an AI model API that you have obtained yourself to use our Services. In that case, we only provide the interaction medium for model calls, including the locally run interactive interface, supporting tools, and communication technologies, while the model service is provided by the model service provider you call.”
“4.We cannot control or guarantee the accuracy of content, service quality, or data privacy and security of any models you connect to on your own, and we assume no liability for their actions. If you use a third-party large model, please carefully read the relevant legal agreements of that model service provider, such as the user agreement and privacy policy.”
“When using the Services, you may obtain and configure third-party plugins, MCP servers, Skills, Hooks, and their dependencies on your own. These tools and services are provided by third parties, and we are not responsible for third-party tools or services.”
“以下为 DeepSeek 已发布的主要模型信息，包含模型名称、发布时间、技术报告与模型卡 V4.0 DeepSeek-V4 New 2026年4月24日 模型卡 技术报告 V3.2 DeepSeek-V3.2 2025年12月1日”
“To learn about the filing status and basic principles of the DeepSeek model and algorithm, please see the Model Principles and Training Methods Disclosure”
?Technical dependency observed: AWSObservation
Technical evidence (DNS, certificate, or HTTP) shows a dependency on AWS as a platform provider. This dependency is not identified in the vendor's published materials.
An undisclosed provider dependency is exactly the gap independent verification exists to surface.
Question for vendor: Verification required — AWS appears to be involved as a platform provider: confirm whether this dependency exists, and whether it processes customer data.
“Official models : When you use the DeepSeek official models configured by default after registering and logging in, we provide you with a full-chain service from the foundation model to the interaction medium.”
“Custom models : You may also configure an AI model API that you have obtained yourself to use our Services. In that case, we only provide the interaction medium for model calls, including the locally run interactive interface, supporting tools, and communication technologies, while the model service is provided by the model service provider you call.”
“以下为 DeepSeek 已发布的主要模型信息，包含模型名称、发布时间、技术报告与模型卡 V4.0 DeepSeek-V4 New 2026年4月24日 模型卡 技术报告 V3.2 DeepSeek-V3.2 2025年12月1日”
“4.We cannot control or guarantee the accuracy of content, service quality, or data privacy and security of any models you connect to on your own, and we assume no liability for their actions. If you use a third-party large model, please carefully read the relevant legal agreements of that model service provider, such as the user agreement and privacy policy.”
“When using the Services, you may obtain and configure third-party plugins, MCP servers, Skills, Hooks, and their dependencies on your own. These tools and services are provided by third parties, and we are not responsible for third-party tools or services.”
Assurance dimensionsweighted components of one score, not eight separate ratings
Organisation & AI governance15% of the score40
Who owns AI risk: policies, responsible-AI principles, human oversight and administrative controls.
Assessed against ISO 42001 · NIST AI RMF · AIUC-1
“（5）为贯彻落实《人工智能生成合成内容标识办法》及《网络安全技术 人工智能生成合成内容标识方法》等国家标准的相关要求，为防止混淆和误认，减少不准确生成信息的潜在影响，DeepSeek已在平台内对人工智能模型生成内容添加标识，并明确提醒用户相关内容由人工智能模型生成。”
“To learn about the filing status and basic principles of the DeepSeek model and algorithm, please see the Model Principles and Training Methods Disclosure”
“Using the Services to engage in or assist high-risk activities, such as military and warfare, the development of weapons, explosives, or hazardous materials, the management or operation of critical infrastructure (such as transportation and energy), the creation or distribution of controlled substances or services, the generation or dissemination of information related to self-harm, and activities with a high risk of economic harm, including gambling, automated determination of credit eligibility, and employment and education decisions.”
AI system15% of the score47
What AI the vendor actually runs and where it sits in the product, the map everything else is judged against.
Assessed against ISO 42001 · NIST AI RMF
“including but not limited to a platform whose core function is calling AI models and tools”
“The Services can access a variety of systems and can execute model-generated code and commands on behalf of the user, load third-party plugins, and access networks, processes, credentials, and files made available to them. Although most foundation models include basic defenses against prompt injection, inherent risks remain when using DeepSeek Harness to interact with the internet or to operate through other untrusted data sources.”
“Official models : When you use the DeepSeek official models configured by default after registering and logging in, we provide you with a full-chain service from the foundation model to the interaction medium.”
“Custom models : You may also configure an AI model API that you have obtained yourself to use our Services. In that case, we only provide the interaction medium for model calls, including the locally run interactive interface, supporting tools, and communication technologies, while the model service is provided by the model service provider you call.”
How AI features are evaluated before and after release, evals, red-teaming, monitoring.
Assessed against NIST GenAI · NIST AI RMF · ISO 42001
“Harness and related technologies such as agents are still at an early stage of development and are innovative and experimental in nature. The DeepSeek Harness Services are also still in a continuous testing phase and must not be regarded as fully secure or ready for production environments.”
“以下为 DeepSeek 已发布的主要模型信息，包含模型名称、发布时间、技术报告与模型卡 V4.0 DeepSeek-V4 New 2026年4月24日 模型卡 技术报告 V3.2 DeepSeek-V3.2 2025年12月1日”
How changes to models and AI features are controlled and communicated: versioning, notice, rollback.
Assessed against ISO 27001 · ISO 42001
“With the continuous iterative development of generative artificial intelligence technologies, models, and products, as well as changes in laws and regulations, we may add, upgrade, change, suspend, or terminate the Services, or make necessary adjustments to the technology, methods, performance, and the like of the Services, and may conduct internal or external testing for newly added service features.”
“we may revise these Terms from time to time. The revised content will form an integral part of these Terms. Once announced, it replaces the original terms. Please refer to the latest version of these Terms on the official website. If you do not accept the modified terms, please stop using the Service immediately. Your continued usage of the Service will be considered as your acceptance of the modified terms.”
AI system description: vendor-evidenced, not yet independently corroborated.
Model10% of the score60
Which model providers and versions process customer data, and how changes to them are disclosed.
Assessed against NIST GenAI · ISO 42001
“Official models : When you use the DeepSeek official models configured by default after registering and logging in, we provide you with a full-chain service from the foundation model to the interaction medium.”
“Custom models : You may also configure an AI model API that you have obtained yourself to use our Services. In that case, we only provide the interaction medium for model calls, including the locally run interactive interface, supporting tools, and communication technologies, while the model service is provided by the model service provider you call.”
“以下为 DeepSeek 已发布的主要模型信息，包含模型名称、发布时间、技术报告与模型卡 V4.0 DeepSeek-V4 New 2026年4月24日 模型卡 技术报告 V3.2 DeepSeek-V3.2 2025年12月1日”
Model provider transparency: vendor-evidenced, not yet independently corroborated.
Customer data15% of the score60
Whether customer data trains models, how long it is retained, where it is processed, and how it is protected.
Assessed against ISO 27018 · OAIC APPs · EU AI Act
“（3）为向您提供连续、高质量的服务， 在经安全加密技术处理、严格去标识化且无法重新识别特定个人的前提下，我们可能会将服务所收集的输入及对应输出，用于模型训练和服务的优化。在此前提下，您同意授予深度求索在全球范围内一项免费的、非排他的许可使用权，以使我们对前述内容进行处理。 如您拒绝将您的数据用于模型训练，可以通过第七条的联系方式联系我们来选择退出，关闭后您的输入和输出不会再用于我们的模型训练。”
“(3)Under the premise of secure encryption technology processing, strict de-identification rendering, and irreversibility to identify specific individuals, we may, to a minimal extent, use Inputs and Outputs to provide, maintain, operate, develop or improve the Services or the underlying technologies supporting the Services. If you refuse to allow us to process the data in the manner described above, you can opt out by contacting us through the contact information in Article VIII.”
“3.Your Input will be sent directly to the corresponding AI model for processing, and the Output will also be returned directly by that AI model. DeepSeek Harness does not store or process your Input and Output on the server side.”
“（1）为履行法定义务的要求，深度求索有权采取技术手段等措施对用户使用本服务的行为及信息进行审查，包括但不限于对输入和输出进行审查、建立风险过滤机制、建立违法内容特征库等。作为本服务的使用者，”
“For behaviors suspected of violating laws and regulations or involving illegal activities, relevant records will be retained, and reports will be made to the competent authorities in accordance with the law, cooperating with their investigations.”
Customer data treatment: vendor-evidenced, not yet independently corroborated.
AI supply chain10% of the score40
The named chain of AI and infrastructure providers behind the product, and how it is disclosed.
Assessed against ISO 27001 · ISO 27017
“Official models : When you use the DeepSeek official models configured by default after registering and logging in, we provide you with a full-chain service from the foundation model to the interaction medium.”
“4.We cannot control or guarantee the accuracy of content, service quality, or data privacy and security of any models you connect to on your own, and we assume no liability for their actions. If you use a third-party large model, please carefully read the relevant legal agreements of that model service provider, such as the user agreement and privacy policy.”
“When using the Services, you may obtain and configure third-party plugins, MCP servers, Skills, Hooks, and their dependencies on your own. These tools and services are provided by third parties, and we are not responsible for third-party tools or services.”
Security foundation15% of the score40
How security problems are found, reported and handled: disclosure channels, bug bounty, incident response.
Assessed against ISO 27001 · NIST AI RMF
“If you discover a security vulnerability, we recommend that you report it privately through the repository at [email protected]”
“Sandboxes, approval prompts, and permission controls can reduce risk, but they do not guarantee isolation and cannot guarantee that harm will be prevented. Even where restrictions are correctly enforced, they cannot protect resources that the project has been permitted to access.”
“We will take necessary measures no less rigorous than industry practice to ensure the cybersecurity and stable operation of the platform, and will strive to enhance and improve our technology to ensure your product and service experience.”
Independent assurance evidence10% of the score0
Whether claimed certifications and attestations are independently corroborated, checked against official registries and issuing bodies. Which certifications, and what each one does and does not establish, is stated per certificate.
Assessed against ISO 17021-1 · ISO 42006 · ISO 27006-1 · ISO 27008
Independent assurance: not publicly evidenced.
Legal & contractual10% of the score60
The contractual backbone a buyer can rely on, terms, DPAs, privacy commitments and regulatory posture.
Assessed against EU AI Act · OAIC APPs
“更新日期： 2026年9月20日 生效日期： 2026年9月27日 亲爱的用户，欢迎您使用DeepSeek Harness 产品及服务！ DeepSeek Harness 产品及服务由杭州深度求索人工智能基础技术研究有限公司（以下简称“我们”或“深度求索”）所有和负责运营。在使用本服务前，请您务必仔细阅读并理解本《DeepSeek”
“1.The establishment, execution, interpretation, and resolution of disputes under these Terms shall be governed by the laws of the People's Republic of China in the mainland.”
“OUR AGGREGATE LIABILITY UNDER THESE TERMS WILL NOT EXCEED THE AMOUNT YOU PAID FOR THE SERVICE THAT GAVE RISE TO THE CLAIM DURING THE 12 MONTHS BEFORE THE LIABILITY AROSE.”
“You may apply the Inputs and Outputs of the Services to a wide range of use cases, including personal use, academic research, derivative product development, training other models (such as model distillation), etc., as long as such usage is legal and adhere to these Terms.”
Legal & contractual transparency: vendor-evidenced, not yet independently corroborated.
Not graded: Agent, not applicable to this scan.
What would strengthen assuranceeach figure is from today's score, resolving one changes the others
Resolving an evidence gap strengthens the dimension it belongs to, and with it the score: 49 → up to 68 with every identified gap resolved.
AI supply chainclick a node to focus it · drag to pan · zoom with the controls
View as list
What you inheritTrustyCyber's AI Shared Responsibility Matrix v1.0
This scan names the vendor’s AI supply chain without assessing the responsibility split, that depends on how you consume the product. Under the common managed models, a deployer inherits 16 of the matrix’s 41 responsibilities from providers, each with named evidence owed.
Map your inherited responsibilitiesAssurance evidence: certifications 1, registry checks and verification ladders, click to view
Not claimed in any public source scanned. This is the AI-management-system certification — in its absence, the vendor’s AI governance rests on its general security and privacy certifications.
Sources 7, click to view
Appendix: standards landscape 8, click to view
Where each standard behind this assessment stands today, from TrustyCyber’s continuously maintained standards radar.
Want to go further?
This scan assesses DeepSeek at an organisational level. Assurance can vary significantly between products, models and agents, enter the product you’re evaluating to run a deeper AI System Assurance Scan.
Scan a specific product →This automated result is based on publicly available information at the time of scanning. It is not an audit, certification, legal opinion or assurance engagement. Publicly unavailable evidence may materially change the conclusion.
Monitor for changes
Get an email if its TRUSTYCYBER Score, grade or certifications change. One credit per refresh; reading the report always stays free.