Thanksgiving DealsAmazon USThanksgiving week: Black Friday picks start earlyAmazon US: current deals, useful picks and tech finds.Check DealsSlow PC?RecommendedPC slow today? Run a repair scan before it gets worseResolve common Windows issues and optimize system performance.Scan NowThanksgiving DealsAmazon USDo not wait for the rush - compare today's dealsAmazon US: current deals, useful picks and tech finds.See Picks×
FREEDOM251

Freedom251 › Blog

Chinese AI Models: Biggest Deployment Risks and How Teams Can Mitigate Them

By Freedom251 Team · · 12 min read

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.

Deploying a Chinese AI model is neither automatically unsafe nor automatically safe. The biggest risks arise from the specific model, its operator, the way it is hosted, the data and tools it can reach, and the organization’s ability to verify and control all of those factors. A sound decision combines country-of-origin review with model-level due diligence, supply-chain checks, data protection, adversarial evaluation, strict access controls, legal review, and a documented exit plan.

The Australian Signals Directorate and Australian Cyber Security Centre (ASD/ACSC) guidance puts the point precisely: “Country of origin should be considered as one element of this assessment and not, by itself, as evidence that a model is trustworthy or untrustworthy.” That principle is especially important when comparing a hosted service with a locally operated open-weight model, because each creates a different set of responsibilities.

As an Amazon Associate I earn from qualifying purchases.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

What are the biggest risks of deploying Chinese AI models?

The material risks fall into six connected categories: confidentiality and jurisdiction, model and software provenance, leakage of sensitive data, unreliable or biased behavior, prompt injection and unsafe tool use, and legal or licensing uncertainty. None is unique to Chinese models; the concern is that jurisdiction, governance, documentation, access arrangements, and export rules may be less familiar or harder for an adopting organization to verify.

Risk area What can go wrong Primary controls
Data and jurisdiction Prompts, outputs or telemetry may be retained, accessed or transferred outside your approved locations. Map data flows, obtain contractual commitments, minimize sensitive inputs, encrypt and restrict access.
Supply chain Weights, conversions, libraries, containers or plugins may be altered, vulnerable or malicious. Approved sources, hashes and signatures, dependency review, sandboxing and provenance records.
Confidentiality Fine-tuning or adapters may retain information that can later be extracted or inferred. Treat checkpoints as sensitive, protect copies, validate datasets and run disclosure tests.
Behavior The model may be inaccurate, biased, ideologically framed, unsafe or prone to unauthorized instruction following. Evaluate the exact version in realistic languages and tasks; add human review and external policy controls.
Connected tools Prompt injection can turn an untrusted document or web page into an unauthorized action or data exfiltration. Least privilege, network egress limits, bounded tools, approval gates and injection testing.
Law and continuity Licenses, privacy obligations, procurement rules or export controls may restrict a planned use or later movement of artifacts. Legal review, written obligations, version inventory and a tested suspension and replacement process.

These risks compound. For example, a converted model from an unknown distributor may run on a locally controlled server, yet still include an unsafe plugin and have no reliable update channel. Conversely, a well-documented model behind a hosted API may have strong operational controls but expose regulated prompts to a provider or jurisdiction that your contracts do not permit.

Popular on Amazon
Sale
Silkland 40Gbps USB C 90 Degree Right Angle Adapter 2 Pack, 240W USB4 | Male to Female Extender, 8K@60Hz, Thunderbolt 4/3, Steam Deck, ROG Ally, MacBook, Switch 2, iPhone 18/17/16/15, Quest 3 VR
  • 【90 Degree Full Function USB C Adapter】Supports the latest USB4 standard for up to 40Gbps of bandwidth, 8K video support, and up to 240W Fast charging; Support USB PD 3.1 Extended Power Range (EPR) including 140W, 180W and 240W. Your productivity is easily doubled. *Warm Tip: For the best fit, please make sure the USB-C port is fully exposed; thick or bulky cases may affect compatibility.
  • 【Reliable Quality】Made from high-quality, scratch-resistant aluminum alloy, delicate metal frame. get excellent wear resistance and heat dissipation. Non-Slip Design for easier insertion and removal.
  • 【Small and portable】Ultra thin size, ultra compact design, a petite physique that matches the minimalist design of your modern laptops. Coming with a slim figure that won't block the port next to the one it's plugged into.
  • 【Compatibility】Widely work with most usb c, thunderbolt 4/3 devices. Compatibilit for ROG Ally, Steam Deck, Switch, MacBook Pro, Quest 3 VR headset, Surface Book 2, iPhone 18/17/16/15, Samsung Galaxy S23/S22/S21, Note 20/10, Google Pixel, ChromeBook, etc.
  • 【Excellence Customer Support】If you experience any issues or cable failure, simply reach out to our dedicated customer service team. We provide a prompt response within 18 hours, providing you with the necessary technical support to resolve any concerns.

Does a Chinese model’s country of origin determine whether it is safe?

No. Origin is a due-diligence factor, not a verdict. Assess the developer, distributor, modifiers, exact version, provenance, documentation, license, jurisdiction and available security disclosures. If those facts cannot be established well enough for the proposed use, reject that use rather than relying on a nationality-based assumption.

Country can influence the legal environment, government access obligations, transparency practices, export controls and ability to obtain remedies. It can also affect where support staff, logs or update infrastructure are located. Those are concrete questions to investigate, not proof that every service or model from one country follows the same practices.

Free tools Windows power users keep installed

One-click scans. No signup required.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Use a written risk record that separates verified facts from assumptions. Record the model identifier and checksum, who supplied each artifact, where inference and support occur, what data is retained, and which entity is contractually responsible for incidents. Revisit the record when the model, provider, serving stack or intended use changes.

What happens to company data when a team uses a hosted Chinese AI service?

It depends on the service architecture and contract. Before sending production data, establish where inference occurs, what prompts and outputs are logged, how long they are retained, who can access them, whether they are used to improve the service, what onward transfers occur, and which security and deletion commitments apply.

Ask for a current data-flow diagram or equivalent technical description. Identify customer-facing APIs, support channels, observability systems, abuse-monitoring systems, backups and subprocessors. “The model is hosted in our region” does not by itself answer where support access, telemetry processing or disaster-recovery copies occur.

Apply data minimization even when contractual assurances are strong. Remove direct identifiers, redact secrets, tokenize records and route only the fields needed for the task. Keep highly regulated or strategically sensitive data out of a service until privacy, procurement and legal teams have approved the arrangement. Log the classification of every dataset sent to the service so an incident investigation can determine what was exposed.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
Also popular on Amazon
Sale
Anker USB C Hub, 5-in-1 USBC to HDMI Splitter with 4K Display | 85W Max PD, 1 Powered USB-C 5Gbps & 2 Powered USB-A 3.0 5Gbps Data Ports, for MacBook Pro, MacBook Air, Dell and More (Black)
  • 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
  • Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
  • Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
  • HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
  • What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.

The reviewed evidence does not establish that every Chinese-hosted service has identical data practices or that every interaction is accessible to a government. State the actual provider’s terms and architecture; where they are unknown, treat that uncertainty as a reason to limit the use.

Is self-hosting a Chinese open-weight model safer?

Self-hosting can keep inference and primary data inside infrastructure your organization controls, but it does not remove risk. It transfers responsibilities that a hosted provider might otherwise perform: hardening, patching, abuse detection, monitoring, update management, incident response and secure disposal.

Decision question Hosted API or service Local open-weight operation
Where do prompts and outputs go? Provider infrastructure, support systems and possibly subprocessors; verify locations and retention. Usually inside your environment, but telemetry, plugins and accidental egress can still leave it.
Who controls versions? Provider may change the model or serving behavior; require notice and pinning options. You choose versions, but must track updates, vulnerabilities and rollback artifacts.
Who runs security operations? Shared responsibility: the provider secures its service while you secure identities, prompts, tools and configuration. Your team owns the full serving stack, monitoring, abuse controls, patching and response.
What must be verified? Provider security disclosures, contracts, data use, access controls and subprocessors. Weights, conversions, libraries, containers, plugins, network paths and host hardening.
How much control is available? Less control over internals, but potentially mature availability and monitoring. More control over data, version and network, with higher operational workload.

Choose the architecture that your team can operate reliably. A local server with unrestricted outbound access, unpatched dependencies and no incident process may be less defensible than a carefully contracted service. Conversely, a hosted endpoint that receives restricted data without an approved transfer mechanism may be unacceptable even if its model quality is strong.

How should a team check model files and dependencies?

Treat the model as a software supply-chain component, not as a harmless data file. Weights can be converted, quantized, merged or fine-tuned by parties other than the original developer, and the serving framework, container, tokenizer, extensions and plugins can introduce additional vulnerabilities.

What’s actually slowing this PC down?

Pick the symptom - the matching free tool is one click away.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
  1. Define an approved source. Record the original developer, distributor and download location. Prefer a channel that publishes version history, security contacts and reproducible release information.
  2. Capture immutable identifiers. Store the exact version, digest or hash, file list, license and download date. Preserve the original artifact so a later replacement can be detected.
  3. Verify authenticity where available. Check digital signatures, signed release metadata and provenance attestations. A matching hash proves that a file has not changed since that hash was issued; it does not prove that the original file is trustworthy.
  4. Review the execution format. Prefer formats that do not execute arbitrary code during loading. Treat serialized objects, custom loaders and conversion scripts as executable software and inspect them accordingly.
  5. Scan the complete stack. Inventory operating-system packages, language libraries, GPU drivers, serving frameworks, containers, extensions and plugins. Track vulnerabilities and license obligations for each dependency.
  6. Load in a quarantine environment. Use a disposable sandbox with no production credentials, no sensitive files and tightly restricted network access. Observe file, process and network activity before the model is admitted to a staging environment.
  7. Test for concealed behavior. Probe unusual prompts, data-exfiltration attempts, instruction-hierarchy conflicts and tool calls. A benchmark score or malware scan alone is not sufficient assurance.

Maintain a software bill of materials or equivalent inventory for every deployment. When a distributor publishes a replacement, a new quantization or an adapter, treat it as a new artifact that requires review rather than as a routine download.

How can teams prevent sensitive-data leakage from training and fine-tuning?

Fine-tuning on internal records can create a model, checkpoint or adapter that contains information derived from those records. Protect these artifacts as sensitive data, even when the training files have been deleted.

Also popular on Amazon
Sale
Anker USB C Hub, 7in1 Multi-Port USB Adapter, 4K@60Hz USBC to HDMI Splitter | 85W Max PD, 2 USB 3.0 & 1 USBC Data Ports, SD/TF Card Reader, for Laptops/Mac and Other Type C Devices (Charger Not Included)
  • Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
  • Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
  • Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
  • Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
  • What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
  • Classify training, validation, prompt and output data before use; document lawful purpose and provenance.
  • Remove secrets and direct identifiers, and use carefully governed synthetic or redacted data where it meets the task.
  • Restrict who can read, copy, export or attach checkpoints and adapters. Separate development, evaluation and production credentials.
  • Encrypt artifacts in transit and at rest, with keys managed separately from the storage system.
  • Use integrity checks and access logging for datasets, checkpoints, tokenizers and configuration files.
  • Test whether memorized records, rare strings or membership information can be extracted or inferred through ordinary and adversarial prompts.
  • Define retention and secure-deletion rules for temporary caches, experiment runs, logs and backups.

Operational data also needs protection. Validate inputs, monitor for tampering and drift, and prevent a user from placing confidential documents into an evaluation or debugging channel that is less controlled than production.

How should a team evaluate bias, reliability and instruction following?

Evaluate the exact model version and configuration in the environment where it will operate. Generic benchmark results do not predict performance on your languages, workflows, user population or connected tools.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Build a test set from realistic tasks and include applicable Chinese, English and other required languages. Measure systematic inaccuracies, refusal behavior, hallucinated citations, unsafe content, sensitive-data disclosure, unauthorized instruction following and resilience to conflicting instructions. Include scenarios involving culturally or politically sensitive subjects when the model will provide advice, make recommendations or generate public content.

ASD/ACSC notes that ideological bias may have limited effect on objective technical work but can be more consequential in judgment, advice, decision support, narrative and content tasks. Set a human-review threshold based on impact: a draft classification may receive sampling review, while a decision affecting safety, employment, finance, health or access should require qualified human approval and a documented fallback.

Keep policy enforcement outside the model. Use application-level content rules, data-loss prevention, identity controls and audit logs because model safeguards can vary by version, configuration or user prompt. Record test cases, expected outcomes, observed failures and the decision to accept, mitigate or reject each risk.

What controls stop prompt injection and unsafe tool use?

A model connected to email, files, databases, browsers or business systems can turn an unsafe response into a real-world action. Untrusted text must never be allowed to redefine system instructions, tool definitions or authorization boundaries.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
Also popular on Amazon
Sale
BENFEI USB C Hub 5-in-1 with 4K HDMI(Certified), 100W Power Delivery, 3 USB-A, Silicone Cable, Aluminum Case Compatible with MacBook Pro/Air, iPad Pro, iMac, iPhone 15 Pro/Pro Max, XPS, Thinkpad
  • Portable and powerful USB-C HUB: BENFEI USB Type-C HUB, with super-soft and knot-free silicone woven design cable, meets most mobile office needs. Compact, lightweight, stylish, and powerful portable USB C Hub equipped with 1 x HDMI port, 1 x 100W charging, and 3 x USB ports. 18-month warranty, 24-hour response, to ensure you feel at ease when using our product.
  • Design centered on comfort and reliability: Thanks to BENFEI's end-to-end in-house cable production capability, in-house PCBA and assembly capability, using the industry's most advanced silicone woven design and process, 20cm cable in length, no knots, super-soft, the HUB is easy to use in all scenarios: laptop, tablet, stand etc. Super-soft, 25000+ life cycles, to meet your daily carrying and office needs.
  • 100W Charging: Support up to 90W USB C pass-through charging via Type-C port to keep your laptop powered. 10W is reserved for other interface operations. No data and video function on the Type-C port.
  • 4K HDMI Display: The HDMI port supports media display at resolutions up to 4K 30Hz, keeping every incredible moment detailed and ultra vivid. Please note that the C port of the Host device needs to support video output.
  • Transfer Files in Seconds: Transfer files and from your laptop at speeds up to 10 Gbps with USB A 3.2 port. Extra 2 USB A 2.0 ports are perfectly for your keyboards and mouse.
  1. Give the model only the minimum tools and permissions required for the task.
  2. Use separate identities for reading and writing; make write, delete, payment and privilege-changing operations unavailable by default.
  3. Restrict outbound network destinations and block direct access to internal metadata services, credentials and management interfaces.
  4. Validate tool arguments in deterministic code, including destination, file path, record scope and quantity limits.
  5. Require a human approval step for irreversible, high-impact or externally visible actions.
  6. Label retrieved text as untrusted data and keep it separate from system policy and tool schemas.
  7. Test indirect prompt injection, data exfiltration, cross-tenant access and attempts to chain harmless tools into a harmful outcome.
  8. Log the user, model version, retrieved content, proposed action, policy decision and final tool result for investigation.

Use a kill switch that can revoke tokens, disable tools and isolate the model without waiting for a model update. A safe response is not enough if the surrounding application still permits an unauthorized action.

What legal, licensing and export issues should be reviewed?

Review the developer and distributor jurisdictions, model license, privacy and data-transfer rules, procurement requirements, sector regulations and export controls for the actual model, weights, data and infrastructure. Open-weight availability does not mean unrestricted commercial use, redistribution, fine-tuning or cross-border transfer.

Licenses can impose attribution, notice, use restrictions or obligations that flow to modified versions. Ask counsel to examine both the model license and the licenses of bundled code, datasets, tokenizers and serving components. Preserve the license text that applied to the exact artifact you approved.

China’s Ministry of Commerce has discussed revising its catalogue of technologies prohibited or restricted from export and addressed overseas deployment and invocation services. The referenced page was not available for full verification in the source review, so treat it as a diligence prompt rather than a settled statement of the final rule or its application. Obtain jurisdiction-specific advice before moving weights, providing remote access, offering an inference service or sending controlled data across borders.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Support on Ko-Fi

What should be in a safe deployment process?

Use a staged process that produces evidence a security, privacy and business owner can review.

  1. Define the use and impact. State what the model may do, who relies on it, what harms are possible and which data classifications are in scope.
  2. Identify the exact asset. Record developer, distributor, version, checksum, license, modifiers, adapters, dependencies and support contacts.
  3. Map architecture and data flows. Document inference locations, logs, backups, telemetry, users, tools, network destinations and subprocessors.
  4. Perform legal and procurement review. Confirm privacy, transfer, export, sector and contractual requirements before production data is used.
  5. Quarantine and inspect artifacts. Verify provenance, scan the stack, sandbox loading and remove unnecessary network access.
  6. Evaluate behavior. Run realistic multilingual, security, bias, leakage, reliability and tool-use tests against the exact configuration.
  7. Add compensating controls. Enforce least privilege, egress restrictions, approval gates, content policies, rate limits and monitoring outside the model.
  8. Stage and observe. Start with low-impact users or synthetic data, review logs and failure reports, and define measurable go/no-go criteria.
  9. Approve with an owner. Assign responsibility for security operations, model changes, incident response and periodic reassessment.
  10. Maintain an exit path. Keep replacement models, exported data, configuration backups and rollback procedures so service suspension does not halt critical work.

How should teams monitor and reassess the deployment?

AI risk changes when the provider changes its terms, the model is updated, a new adapter is installed, a tool is added, a vulnerability is disclosed or the data and user population expand. Keep an inventory of versions, artifacts, adapters, dependencies, prompts, policies and configurations.

Also popular on Amazon
𝟒𝐏𝐚𝐜𝐤 𝐔𝐩𝐠𝐫𝐚𝐝𝐞𝐝 USB to USB C Adapter,Type-C Female to A Male Charger Converter for iPhone 18 Charger,iWatch Ultra SE 9 10,iPhone 17 Pro Max 16 15 14,Samsung S25,AirPods,iPad,Car,Xbox
  • 【Read before You Buy】 Note: This adapter is USB 2.0 version and cannot transmit HDM, VGA, or other video signals. Nevertheless, they are fully capable of charging devices,duplicating files, and can be seamlessly integrated with USB-C earphones, wireless network adapters, and SD card adapters.
  • 【Easy to Use and Safe】Thanks to modern manufacturing techniques, we've made these adapters sufficiently compact so that they can remain inserted in your device without noticeable protrusion or inconvenience. This means you don't have to constantly plug and unplug them, reducing the likelihood of wear and misalignment, and making them more convenient to carry around.
  • 【Reliable Quality】Engineered for enhanced durability, durable aluminum alloy shell material,get excellent wear resistance and heat dissipation,the male USB C and female USB connectors & the male USB and female USB C connectors are designed to be plugged and unplugged up to 20000 times. Provides data transfer, device charging, memory space expansion, and additional computer peripherals.
  • 【Think more for you】Do you need to connect a USB-C-enabled device to an odd device that only has USB-A ports? Then these USB-A male to USB-C female adapters are made for you.Simply insert the adapter into the USB-A port on the device, and you can instantly connect your USB-C device to the device that originally didn't have the correct connection port.
  • 【Trusted xiwxi Quality and Service】As a responsible brand seller, you will receive xiwxi's 12-month Quality assurance and lifetime friendly customer service, you're not just buying a product,you're investing in peace of mind with all-round quality and exceptional customer service.

Set review triggers rather than relying only on an annual calendar. Reassess after material model or serving-stack changes, significant incidents, new data classifications, new jurisdictions or changes in tool permissions. Monitor access, unusual prompt patterns, extraction attempts, policy violations, tool calls, egress and model-quality drift.

Maintain an incident plan with contacts for the provider, infrastructure, privacy, legal and business owners. It should specify how to revoke credentials, disable tools, isolate hosts, preserve evidence, notify affected parties and restore service with a known-good version. If significant risks cannot be controlled, suspend or retire the deployment instead of accepting an indefinite exception.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

What is the practical decision rule?

Adopt a Chinese AI model only for a defined use where the organization can verify provenance, understand data handling, satisfy legal obligations, test behavior in context and operate the required controls. Hosted use demands provider and shared-responsibility scrutiny; self-hosting demands stronger internal security operations. If key facts remain unknowable or the consequences of failure exceed the available controls, choose a different model or do not deploy for that use.

Frequently Asked Questions

Can local deployment guarantee that prompts never leave the company?

No. Telemetry, update checks, remote support, plugins, cloud-managed GPUs and misconfigured proxies can still create outbound paths. Verify and monitor every network connection, then enforce an allowlist and test the host with production-like controls before sending sensitive prompts.

How often should an AI model risk assessment be repeated?

Repeat it whenever the model, weights, adapter, serving framework, provider terms, tool permissions, data classification or deployment geography changes materially. A scheduled review can supplement those triggers, but it should not replace change-based reassessment.

Are Chinese AI models prohibited for government or regulated work?

There is no single worldwide rule covering every model and use. Eligibility depends on the agency, sector, contract, jurisdiction, data location, license and applicable export or procurement controls, so obtain the relevant legal and procurement determination for the specific deployment.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

What evidence should an approval board request from a model supplier?

Request versioned model cards, security contacts, vulnerability-notification procedures, data-retention and access terms, subprocessors, update and rollback practices, license text, provenance information and incident-notification commitments. Missing documentation should be recorded as a risk rather than silently assumed away.

Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply. Prices as of November 20, 2026 12:44 pm GMT+0530.

Leave a Reply

Your email address will not be published. Required fields are marked *

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Recommended PC Tool
Recommended PC Tool
Windows Errors? Fix Them Before They SpreadFree repair scan
Crashes, No Sound, or Screen Glitches?Free driver scan

Two free Windows tools

One Free Minute Could Fix That PC

Before you go - each of these free tools takes about a minute and tackles what quietly slows a Windows PC down.

Special offer. View Outbyte info, uninstall instructions, EULA, and Privacy Policy.