LinkedIn

Summary of Important Changes

1. New express permission to use data for AI training

The revised policy expressly states that LinkedIn may use personal data to:

• “Develop and train artificial intelligence (AI) models”
• Develop, provide, and personalize its services
• Generate insights using AI, automated systems, and inferences

This is a significant expansion or clarification of permitted data use. The term “personal data” is broad and may include profile information, posts, messages, calendar information, job applications, usage data, inferred characteristics, and data obtained from third parties.

Key risks

• No clear limitation on the types of data used for training. The policy does not clearly exclude private messages, resumes, job applications, customer-provided data, or sensitive personal data.
• No explicit AI-training opt-out is stated. Although the policy provides general privacy settings and rights to object or restrict processing, it does not identify a specific control for preventing data from being used to train AI models.
• No deletion commitment for trained models. Closing an account or deleting data may not remove information already incorporated into training datasets, model parameters, derived insights, or outputs.
• Potentially broad purposes. Training appears tied not only to LinkedIn’s services but also to development of products, services, and related insights.

2. Expanded sharing with affiliates, including Microsoft

The policy states that LinkedIn may share:

• Publicly shared content with affiliates, including Microsoft, to provide or develop their services; and
• Personal data with affiliates to improve, provide, or develop advertising services.

This may permit LinkedIn data to be used across Microsoft-affiliated products and AI systems, subject to the European Regional Privacy Notice where applicable. The policy does not clearly limit such sharing to LinkedIn’s own models or services.

3. Increased use of automated analysis

LinkedIn may use automated systems to:

• Scan messages, where settings allow, for scheduling, drafting responses, summarization, and recommendations;
• Moderate content and detect policy violations;
• Make inferences, including about age, gender, interests, seniority, compensation, and career opportunities;
• Rank content and recommend jobs, contacts, and learning materials.

This creates risks of privacy intrusion, inaccurate profiling, bias, and consequential automated decisions in employment or recruiting contexts.

4. Broader data sources and retention

The policy confirms collection from calendars, contacts, partners, employers, applicant-tracking systems, Microsoft services, advertising partners, cookies, and off-site activity. LinkedIn generally retains data while an account remains open, including data generated or inferred from use.

5. Practical implications

Customers should review:

• Whether confidential business, employee, recruiting, or customer data is uploaded;
• Available AI, personalization, advertising, and message-scanning settings;
• Contracts or regional notices that may impose additional restrictions;
• Whether deletion or objection rights adequately address AI training and derived data.

Executive Summary

The diff appears to contain a wholesale re-rendering or replacement of the policy text, but the substantive wording is almost entirely unchanged. The identifiable changes are primarily formatting, punctuation, and typography corrections. No new substantive permission to use customer data for AI training appears to have been added in this diff.

Identified Changes

1. AI model training language — No apparent change

The policy states that LinkedIn may use personal data to:

• “improve, develop, and provide” products and services;
• “develop and train artificial intelligence (AI) models”; and
• develop, provide, and personalize services and generate insights using AI, automated systems, and inferences.

This language appears in both the deleted and replacement text. Accordingly, the diff does not appear to expand or narrow LinkedIn’s AI-training rights.

2. Existing AI-related risk remains significant

Although not newly introduced, the retained wording creates broad potential uses of customer data:

• It does not clearly limit AI training to de-identified, aggregated, or publicly available data.
• It does not identify specific data categories excluded from training, such as private messages, resumes, employment information, calendar data, or sensitive profile information.
• It does not clearly distinguish between training LinkedIn’s own models and training models operated by affiliates, including Microsoft.
• The policy separately permits use of communications by automated systems for message suggestions, moderation, security, and similar functions.
• The policy also permits sharing public LinkedIn content and personal data with affiliates, including Microsoft, for service and advertising development, subject to the European Regional Privacy Notice.

Customers should therefore treat the AI-training authorization as continuing and potentially broad, even though it is not expanded by this particular diff.

3. Minor wording and formatting corrections

The visible changes include:

• “LinkedIn's” changed to the typographically equivalent “LinkedIn’s.”
• “Learn More.” changed to a period following the apparent removal of a hyperlink.
• “Table of ContentsData” corrected to “Table of Contents.”
• Apparent cleanup of headings, punctuation, and document formatting.

These changes do not appear to alter substantive legal rights or obligations.

Overall Risk Assessment

Change-related risk: Low. The identifiable amendments are largely editorial.

Policy-level data-use risk: Moderate to high. The unchanged policy continues to authorize broad use of personal data—including potentially inferred data and communications—for service development, AI model development and training, personalization, research, advertising, security, and sharing with affiliates. The practical effect of those permissions may depend on settings, regional notices, and applicable law.

Summary

The diff provided only states that approximately 6,625 words were removed. It does not identify which provisions were deleted, nor does it show any additions or replacement language.

AI Training and Customer Data

• No specific changes can be determined regarding:
• Whether the customer’s data may be used to train, fine-tune, test, or improve AI models;
• Whether customer data may be combined with other data for model development;
• Whether prompts, outputs, usage data, telemetry, or derived data may be retained or used for training;
• Whether the customer can opt out of AI training;
• Whether data is anonymized, de-identified, or otherwise protected before use;
• Ownership of models, training outputs, or data-derived insights; or
• Restrictions on human review or access to customer data.

Potential Legal and Commercial Risk

The removal of 6,625 words could materially change the agreement, but the nature and direction of the risk cannot be assessed without the deleted text and any replacement language. Depending on what was removed, the changes could affect:

• Confidentiality and permitted-use restrictions;
• Data protection, security, and breach obligations;
• Retention and deletion requirements;
• Intellectual-property ownership and license rights;
• Liability caps, indemnities, and warranties;
• Audit, compliance, and regulatory obligations;
• Customer termination and data-export rights; and
• Restrictions or disclosures concerning AI systems.

If provisions limiting use of customer data, requiring deletion, or prohibiting model training were removed, the customer could face broader vendor rights to use its data, reduced confidentiality protections, or less control over downstream AI development. Conversely, removal of vendor rights or disclaimers could increase the vendor’s obligations.

Required Information

A meaningful legal analysis requires either:

1. The full before-and-after contract text;

2. A redline showing the deleted and added language; or

3. At minimum, the specific deleted provisions and any replacement provisions.

On the current information, no definitive conclusion can be drawn about changes to AI-training rights or other contractual obligations.