Draw.io

Summary of Important Changes

1. New AI-assisted classification of organisations

• The revised wording introduces a process for determining whether a Google Workspace domain belongs to a school, college, or university that receives draw.io for free.
• The service will use:
• The domain name;
• A list of universities; and
• Otherwise, the domain’s public homepage, which it will provide to an AI model for a classification decision.
• Cloudflare is stated to operate the AI model or related service. The company will store the model’s “verdict” for the domain.

AI-training risk

• The text does not say whether information sent to the AI model is:
• Used to train or improve the model;
• Retained by Cloudflare;
• Used only for this customer-specific purpose; or
• Subject to deletion, confidentiality, or a data-processing agreement.
• Although the source is generally a public homepage, the domain may identify a customer organisation and could indirectly identify individuals or sensitive institutional information.
• The wording should expressly prohibit use of customer-provided data, domain information, or retrieved webpages to train general-purpose AI models unless separately authorised. It should also identify Cloudflare’s role, retention period, location, security measures, and subprocessors.

2. Broader and more detailed usage measurement

The revised policy changes the measurement process substantially:

• It counts Google Workspace accounts only, excluding:
• Personal Gmail and other major personal email providers;
• Google accounts using a company email address without Google Workspace; and
• Educational users at schools, colleges, and universities.
• It counts users who open a diagram, save one, or sign in, by domain and month.
• It creates a keyed-code summary based on each user’s account ID. The text says the summary does not store users’ identities and that the codes are deleted when the month ends.
• However, the revised wording also says the server keeps a monthly count and daily totals for 90 days, while organisational usage data is retained for 24 months. The exact relationship between these datasets is unclear.

3. Additional data sources and disclosure

• The application compares the domain of the user’s Google Workspace account with the domain of the user’s email address.
• Domain information, browser name/version, and historical sign-in information may be processed.
• The application stores settings and a hidden file in the Google Drive app-data folder. The file contains no diagram content and can be deleted by the user or administrator.
• A transitional disclosure states that email domains were sent to the server until October 2026, with the older records retained for 24 months.

4. Key legal and compliance concerns

• Clarify whether keyed account identifiers are personal data or pseudonymous data.
• Define deletion periods consistently, particularly the 24-month and 90-day periods.
• Document the lawful basis, notice, and user/administrator controls for domain and account-level measurement.
• Add explicit AI-use restrictions and vendor obligations covering training, retention, confidentiality, security, and deletion.

Summary

The provided diff states only: “Removed approximately 24 words from the document.” The actual deleted language is not included.

AI Training and Data Use

• No conclusion can be drawn about whether the document’s provisions concerning:
• use of customer data to train, fine-tune, or improve AI models;
• use of prompts, inputs, outputs, or usage data;
• retention or disclosure of customer data;
• human review or monitoring; or
• opt-out, consent, or ownership rights

have changed.

• The deletion of approximately 24 words could materially alter these rights or obligations, particularly if it removed qualifiers such as “aggregated,” “de-identified,” “with consent,” “for service improvement,” or “not used to train models.”

Potential Legal Risks

Because the deleted text is unavailable, the following risks cannot be assessed:

• Whether the provider gained broader rights to use customer data.
• Whether customer consent or an opt-out right was removed.
• Whether restrictions on AI training apply only to identifiable data, rather than all customer data.
• Whether confidentiality, security, retention, or deletion obligations were narrowed.
• Whether the provider can use data for commercial model development or share it with affiliates or third parties.
• Whether the customer remains the owner of inputs and outputs, or whether the provider received expanded rights.

Required Information

To perform a meaningful legal analysis, the actual redlined language is needed, including:

1. The deleted 24 words.

2. The surrounding paragraph or clause.

3. Any replacement language, if applicable.

4. The document version or section heading.

Without the underlying text, the only reliable conclusion is that a potentially important provision was shortened, but the legal effect cannot be determined.

Summary of Important Changes

AI Use and Customer Data

• The revised statement expressly identifies Anthropic and OpenAI as providers of the AI models used for the service’s AI features.
• It also states that Cloudflare handles AI-feature requests, including processing through the nearest data centre.
• The changes do not expressly say whether customer prompts, diagrams, outputs, or other customer data are used to train, fine-tune, or improve Anthropic’s, OpenAI’s, or draw.io’s AI models.
• The statement says providers process personal data under a data-processing agreement and “only on our instructions,” which may limit provider reuse, but this language does not clearly address:
• model training or fine-tuning;
• retention by AI providers;
• human review;
• use of prompts or outputs to improve services; or
• whether customer data is excluded from provider training by contract.
• The security language now expressly covers admin-console data in addition to diagram and technical data, and states that diagram data sent to servers for temporary processing will be publicly documented. This may include data processed by AI features, but the scope and duration of that processing remain unclear.

Expanded Data Collection

• Google Drive licensing analytics are rewritten. The company now counts users associated with a company email domain when they open or edit a diagram or sign in to Google Drive. It stores the number of users per domain and month for 24 months.
• Personal Gmail accounts are excluded from counting. A hidden app-data file is used to prevent duplicate monthly counting; users can delete it through Google Drive settings.
• New admin-console data may include:
• organisation name and domains;
• settings;
• administrators’ passkey public keys;
• passkey names, potentially including personal names;
• sign-in and change activity logs retained for two years; and
• invitation email addresses retained as keyed hashes until used or expired.
• Google account information is checked during administrator sign-in but is not retained.

Changes to Special-Category Data

• The structure of the special-category provisions is substantially reorganised, but the categories remain broadly similar: racial or ethnic origin, political opinions, religion or philosophy, trade-union membership, genetic and biometric data, health data, and sex life or sexual orientation.
• The statement continues to say that draw.io does not deliberately collect such data and prefers users not to provide it. If received, it may be deleted or retained only under specified legal bases.

Other Material Changes and Risks

• Contact-support data may be deleted if the user does not provide enough identifying information; draw.io may then be unable to respond.
• The listed legal entity is clarified as draw.io Limited, an English company, with a registered office and company number.
• Privacy-rights requests now have a specified address: privacy@draw.io.
• The revised wording expands reliance on publicly available and third-party sources to verify or supplement personal information.

Key customer risk: AI processing is now expressly disclosed, but the policy remains ambiguous on whether customer data is used for AI-model training. This should be clarified contractually before submitting confidential diagrams, prompts, or personal data to AI features.

Summary

The diff only states that approximately 24 words were removed, but it does not identify which words, clauses, or sections were deleted.

Legal and Commercial Impact

• The legal effect cannot be assessed without the actual deleted language.
• It is not possible to determine whether the deletion changes:
• Customer or provider obligations;
• Liability, indemnity, confidentiality, or security protections;
• Data ownership or permitted data uses;
• Termination or deletion requirements;
• Regulatory compliance commitments; or
• Service levels, warranties, or payment terms.

AI Training and Data Use

The provided diff does not show whether any language concerning the use of customer data to train AI models was added, removed, or modified.

In particular, the available information does not establish whether the document now:

• Permits or prohibits training models on customer data;
• Allows use of customer data for improving products or services;
• Distinguishes between customer content, usage data, metadata, or anonymized data;
• Requires customer consent or provides an opt-out;
• Limits retention or requires deletion of training data;
• Restricts human review or disclosure of data; or
• Allocates ownership of model outputs, derivatives, or trained models.

Required Information

To perform a meaningful legal analysis, provide the actual 24 deleted words, ideally with the surrounding clause or the original and revised text. Without that information, no reliable conclusion can be drawn about changes to customer-data rights or AI-model training.

Summary

The provided diff only states that approximately 21 words were removed. It does not identify the deleted language or show the surrounding provision.

Key Legal Implications

Because the actual text is unavailable, it is not possible to determine whether the changes affect:

• Customer data ownership or licensing rights
• Provider rights to access, retain, disclose, or commercially use customer data
• Confidentiality or data-security obligations
• Data deletion, return, or retention periods
• Use of customer data to train, fine-tune, evaluate, or improve AI models
• Whether customer data may be used for generalized or shared model training
• Opt-out, consent, or notification requirements
• Allocation of liability for unauthorized use or disclosure

AI Training Risk

No conclusion can be reached about changes to AI-model training. The deleted 21 words could potentially have:

• Removed a restriction on using customer data for training;
• Removed an exception permitting training or service improvement;
• Narrowed or expanded the definition of “Customer Data”;
• Eliminated an opt-out or consent requirement; or
• Changed whether de-identified, aggregated, or derived data may be used.

The deletion could therefore be legally significant, but its effect cannot be assessed without the actual wording.

Information Needed

Please provide either:

1. The full redline showing the 21 deleted words; or

2. The original and revised versions of the affected clause, preferably with surrounding text.

Without that information, any assessment of the change would be speculative.