Online Content, Platforms and Intermediary Liability

Introduction and context

Digital platforms have become essential infrastructures for communication, commerce, education, employment and political participation. Search engines rank information; social-media services host speech and audiovisual material; marketplaces connect buyers and sellers; messaging applications distribute private and public content; cloud and hosting providers supply technical capacity. Their social importance has produced a difficult legal question: when should an intermediary be responsible for content created, uploaded or transmitted by another person?

The answer is neither complete immunity nor automatic publisher liability. Modern regimes attempt to balance four interests: protection from unlawful or harmful content, incentives for platforms to manage risk, preservation of freedom of expression and privacy, and the practical reality that platforms process enormous volumes of user activity. The balance is expressed through safe-harbour rules, notice-and-action procedures, transparency obligations, systemic-risk duties, due-process safeguards and sector-specific regulation.

This chapter uses an internationally comparative approach. Rules differ substantially across jurisdictions and change frequently. Students and practitioners must therefore verify the current text, guidance and judicial interpretation applicable to the relevant country, service and content category. A platform may be governed simultaneously by privacy, consumer-protection, intellectual-property, criminal, telecommunications, competition, accessibility and child-protection law.

Learning objectives

By the end of this chapter, a learner should be able to:

1. Distinguish platforms, intermediaries, publishers, users and content creators.
2. Explain the logic and limits of intermediary safe harbours.
3. Analyse harmful, illegal and misleading content using a rights-based risk framework.
4. Design a defensible content-governance and moderation process.
5. Evaluate platform duties under major comparative regulatory models.
6. Address child safety, accessibility, privacy, transparency and freedom of expression.
7. Apply these principles to a realistic platform-governance problem.

1. Core concepts and terminology

An intermediary is an entity that facilitates interaction or access between third parties. The term may include internet-access providers, hosting services, search engines, social networks, app stores, marketplaces, domain-name services and content-distribution networks. A platform is a broader socio-technical service that combines infrastructure, interfaces, rules, ranking systems and commercial arrangements to enable interactions or content distribution.

A user-generated-content service allows users to create, upload, share or otherwise make content available. A publisher traditionally exercises editorial control over content before publication and may be liable under ordinary publication law. The boundary is not fixed: a service that merely stores content may receive stronger protection than one that materially edits, promotes or monetises it, although recommendation alone does not automatically make a service the legal publisher in every jurisdiction.

Illegal content violates a legal rule. Examples include child sexual abuse material, terrorist recruitment material where prohibited, unlawful threats, copyright infringement, fraud, non-consensual intimate imagery and certain forms of hate speech. Harmful content may cause serious social or individual harm without being unlawful in every jurisdiction, such as self-harm promotion, abusive but lawful speech, dangerous health misinformation or age-inappropriate material. This distinction matters because removing illegal content may be mandatory, whereas restrictions on lawful but harmful content require stronger justification and careful policy design.

Content moderation is the combination of automated tools, human review, user reporting, trusted-flagger programmes, policy rules, sanctions, appeals and enforcement operations used to govern content. Content governance is broader: it includes the design of policies, product architecture, business incentives, institutional accountability, transparency, audits and engagement with affected communities.

Intermediary liability concerns legal responsibility for third-party content or conduct. Liability may arise from knowledge, participation, failure to comply with a valid order, negligence, breach of a statutory duty, consumer deception, copyright infringement, privacy violations or the platform's own representations. Safe harbour is generally a conditional protection, not a universal exemption.

2. Why intermediary liability is difficult

Platforms operate at a scale and speed that make prior human review impossible for most services. A major service may receive millions of uploads daily in many languages. Automated detection can identify known patterns but struggles with context, satire, coded language, reclaimed slurs and rapidly changing events. Human reviewers can interpret context but may experience psychological harm, make inconsistent decisions or lack local knowledge.

The same content can have different meanings in different contexts. A photograph may document a war crime, violate a nudity policy, or be intimate material shared without consent. A political claim may be opinion, satire, propaganda or deliberate deception. A platform therefore makes decisions under uncertainty, while errors have asymmetric consequences:

• False positives remove lawful material, suppress journalism or silence minority groups.
• False negatives leave unlawful or dangerous material available.
• Over-removal may be encouraged by severe penalties and vague definitions.
• Under-removal may result from weak enforcement, inadequate resources or engagement-driven business models.

The law must also account for platform power. A large service can influence elections, markets, reputations and access to information. Yet excessive legal pressure may cause smaller services to exit, concentrate the market in large firms, or privatise decisions about public discourse without adequate due process.

3. Safe-harbour regimes

Safe harbour generally protects an intermediary from some liability for third-party content if specified conditions are met. The policy rationale is to encourage innovation and information exchange while requiring reasonable cooperation against unlawful activity.

3.1 United States: Section 230 and copyright rules

Section 230 of the United States Communications Decency Act generally protects providers and users of interactive computer services from being treated as the publisher or speaker of information provided by another information-content provider. It also contains protection for certain good-faith restrictions on objectionable material. The protection is not unlimited: it does not generally eliminate liability for a platform's own content, intellectual-property claims, federal criminal law, communications privacy law or certain sex-trafficking-related claims. Judicial interpretation and statutory exceptions must be examined for the specific dispute.

Copyright liability is addressed separately through the Digital Millennium Copyright Act, especially its notice-and-takedown framework for eligible service providers. Protection depends on conditions such as designated-agent procedures, repeat-infringer policies and response to compliant notices. A platform should not assume that a general safe harbour covers every legal claim.

3.2 European Union: conditional hosting protection and systemic governance

The European Union's Digital Services Act, applicable in stages from 2023, preserves conditional exemptions for certain intermediary services while creating due-diligence obligations. It distinguishes intermediary services such as mere conduit, caching and hosting, and imposes progressively stronger duties according to service type and scale. Hosting services must generally provide notice-and-action mechanisms, reasons for moderation decisions and internal complaint systems. Very large online platforms and search engines face additional obligations concerning systemic risks, transparency, researcher access and independent auditing.

The DSA does not require general monitoring of all information. It combines enforcement against illegal content with procedural safeguards and systemic accountability. It also addresses dark patterns, advertising transparency, protection of minors and crisis-response mechanisms. The important conceptual shift is from a narrow immunity question toward ongoing organisational and systemic responsibility.

3.3 India: conditional protection and due diligence

Section 79 of India's Information Technology Act, 2000 provides conditional protection to intermediaries for third-party information, subject to statutory requirements and due diligence. The Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021, as amended, prescribe obligations that vary by intermediary category and may include grievance redressal, rules and regulations, compliance personnel, reporting and cooperation duties. Constitutional rights, judicial decisions and procedural requirements remain relevant, especially where governmental removal directions affect speech or privacy.

3.4 United Kingdom and other models

The United Kingdom's Online Safety Act adopts a risk-based framework for regulated user-to-user and search services, with duties concerning illegal content, children and platform systems. It does not simply reproduce a traditional safe harbour. Australia, Singapore, Canada and other jurisdictions use combinations of removal orders, codes, regulator powers, online-safety duties and sector-specific rules. A multinational platform therefore needs a legal mapping exercise rather than a single global moderation policy.

3.5 Conditions commonly attached to protection

Across regimes, recurring conditions include:

A platform should maintain evidence of how these conditions are met: notice records, decision logs, reviewer guidance, escalation records, appeal outcomes and technical controls. Documentation is both a compliance asset and a means of detecting inconsistent enforcement.

4. A risk-based content-governance model

A mature programme separates five questions:

1. What is the content? Identify the material, context, speaker, target, distribution and possible authenticity.
2. What legal rule applies? Determine whether the content is criminal, civilly unlawful, regulated, contractually prohibited or lawful.
3. What harm is foreseeable? Consider severity, likelihood, scale, vulnerability of affected persons and reversibility.
4. What intervention is proportionate? Options include no action, friction, warning, reduced recommendation, age-gating, demonetisation, removal, account restriction or referral to authorities.
5. What process is due? Provide notice, reasons, evidence preservation, appeal and escalation where appropriate.

A useful governance matrix is qualitative rather than falsely precise:

The response must account for context. A documentary or academic record may justify retaining material under controlled access even when the same material would be prohibited as entertainment. Policies should define exceptions transparently and prevent exceptions from becoming arbitrary.

5. Content moderation as an operational system

Moderation consists of interlocking layers:

• Prevention: community standards, user education, identity and age signals, upload friction and secure defaults.
• Detection: hash matching, classifiers, keyword systems, behavioural signals and trusted reports.
• Decision: trained reviewers apply policy with contextual evidence.
• Action: content-level and account-level measures are selected according to severity and recurrence.
• Appeal: the user may challenge factual or policy errors.
• Oversight: quality assurance, audits, external review and governance committees examine outcomes.

Automation should support, not conceal, decision-making. A classifier may be highly accurate on known child-abuse hashes yet unreliable for political misinformation. Performance must be assessed by category, language, demographic group and content format. A single aggregate accuracy figure can conceal serious disparities.

A platform should distinguish content moderation from ranking and recommendation. Leaving content online does not mean it must be algorithmically amplified. Demotion, removal from recommendations, limits on forwarding and contextual labels may reduce harm while preserving access. However, ranking interventions also affect speech and should be transparent, consistent and appealable.

6. Misinformation and disinformation

Misinformation is inaccurate information shared without a demonstrated intent to deceive; disinformation is deliberately deceptive or manipulative information; malinformation uses genuine information in a harmful context, such as publishing private data to intimidate a person. Intent is often difficult to prove, so policies should avoid relying exclusively on inferred motive.

An effective response is layered:

1. Support reliable information and authoritative sources.
2. Add context and provenance information.
3. Reduce virality through forwarding limits or recommendation controls.
4. Correct demonstrably false claims, especially where imminent harm is likely.
5. Remove content that violates a defined rule, such as fraudulent impersonation or incitement.
6. Disclose political advertising, targeting and synthetic media where required.

The freedom-of-expression problem is acute. Governments and platforms can incorrectly label dissent as false, particularly during emergencies. Policies should therefore specify evidence standards, distinguish fact from opinion, permit good-faith uncertainty and provide rapid appeals. Platforms should publish enforcement data without exposing personal information or enabling evasion.

7. Freedom of expression, privacy and procedural justice

Freedom of expression protects democratic participation, artistic work, journalism and unpopular views, but is not absolute. Restrictions should have a lawful basis, pursue a legitimate objective and be necessary and proportionate. Private platforms may have contractual authority beyond constitutional duties, yet their decisions can have public consequences and should respect human-rights principles.

A defensible moderation process includes:

• clear, accessible rules;
• consistent treatment of comparable cases;
• notice of the alleged violation;
• reasons stated in understandable language;
• an opportunity to appeal to a competent reviewer;
• independent or senior escalation for high-impact decisions;
• special safeguards for journalists, researchers and human-rights defenders;
• retention and disclosure controls for evidence;
• protection against retaliation and doxxing.

Privacy must be integrated into moderation. Collecting government identity documents or private messages may improve some controls but creates surveillance, breach and exclusion risks. Data minimisation, purpose limitation, access controls, encryption and defined retention periods are essential. Platforms should not disclose user data to private complainants or governments without a valid legal basis and appropriate process, subject to emergency exceptions defined by law.

8. Child protection and age-appropriate design

Children face distinctive risks: sexual exploitation, grooming, cyberbullying, addictive design, harmful challenges, targeted advertising, exposure to pornography and inappropriate data collection. Child protection must not become a pretext for indiscriminate surveillance or removal of young people's access to information.

Key controls include private-by-default accounts, restricted direct messaging, detection and reporting pathways for grooming, rapid removal of exploitation material, trusted adult and child-friendly complaint channels, age assurance proportionate to risk, limits on profiling and advertising, and safety-by-design testing. Age assurance may use self-declaration, device signals, credentials or estimation technologies; each has accuracy, privacy and exclusion trade-offs.

Platforms should design for the child's evolving capacity. Explanations, consent flows and appeals must be understandable. Staff handling child-safety reports need specialist training, and evidence must be preserved and disclosed to competent authorities according to law. Organisations should conduct child-rights impact assessments rather than treating children solely as a compliance category.

9. Accessibility and inclusion

Content governance can discriminate against people with disabilities or minority languages. Automated systems may misread speech impairments, captions may be inaccurate, and inaccessible appeals may deny a remedy. Compliance should cover both the platform interface and the content process.

Good practice includes keyboard navigation, screen-reader compatibility, captions and transcripts, audio description where appropriate, accessible reporting forms, plain-language notices, alternative formats and human review for high-impact automated decisions. Moderation teams need language and cultural competence. Translation systems should not be the sole basis for removal in legally sensitive cases. Accessibility should be tested with users with disabilities and measured through service-level indicators, not treated as a one-time technical certification.

10. Platform accountability and corporate governance

Accountability operates at several levels:

• Policy accountability: rules are clear, lawful and publicly available.
• Operational accountability: staff, tools and escalation routes can execute them.
• Technical accountability: systems are tested, secure and monitored for bias.
• Managerial accountability: senior leaders own risk, resources and legal compliance.
• External accountability: regulators, courts, auditors, researchers and users can scrutinise decisions.

A board or executive risk committee should receive regular reports on high-severity incidents, appeal reversals, enforcement disparities, government requests, child-safety metrics, accessibility failures and emerging threats. Metrics should not reward removal volume alone. More meaningful measures include substantiated violation prevalence, time to action, appeal reversal rate, repeat-harm rate, language coverage and the proportion of decisions with adequate reasons.

Algorithms used for recommendation, moderation or advertising should undergo impact assessments. The assessment should document purpose, data sources, known limitations, affected groups, human oversight, testing results, incident response and retirement criteria. Independent auditing is valuable only when auditors receive sufficient access and independence; a purely self-certifying audit can become public relations rather than assurance.

11. Applied case: the Meridian marketplace incident

The following is an illustrative fictional case.

Background

Meridian is a multinational marketplace and social-commerce platform. Sellers can upload product listings, livestream demonstrations and communicate with buyers. Its business model rewards engagement and rapid transaction completion. Meridian operates in jurisdictions with different rules on counterfeit goods, consumer fraud, political advertising and online safety.

Problem

During a public-health emergency, coordinated seller accounts promote a counterfeit medical device. The listings use manipulated testimonials, target older users and include a link to a payment page outside Meridian. A user reports the listings, but automated review classifies the material as ordinary health-related commerce. A journalist later identifies hundreds of accounts. Some legitimate listings were removed after a keyword sweep, including a hospital's educational page.

Analysis

The case involves counterfeit goods, consumer deception, possible fraud, privacy and a foreseeable health risk. Meridian is not merely a passive host: it provides ranking, payment integration, seller verification and promotional tools. Those facts do not automatically establish liability, but they increase the importance of reasonable due diligence and accurate representations to users.

The keyword sweep produced over-removal because it treated a term as conclusive rather than evaluating seller identity, claims, evidence and transaction behaviour. The platform also lacked an effective escalation path from a credible journalist report to specialist review. Its terms promised safe shopping but did not explain how health claims were verified.

Approach and solution

Meridian should immediately suspend the identified network, preserve relevant evidence, notify affected purchasers, block the external payment redirection and refer suspected criminal conduct to competent authorities where legally required. It should refund or protect consumers under its marketplace policy and review whether its own promotional systems amplified the listings.

For prevention, Meridian should introduce risk-based verification for health products, require substantiation for high-risk claims, limit recommendation until verification, deploy graph analysis for coordinated accounts, and create a specialist escalation channel. Automated signals should identify candidates, while trained reviewers assess context. Legitimate educational material should be exempted through an evidence-based exception rather than a broad keyword exclusion.

Users should receive reasons for removal and a route to appeal. The company should publish a post-incident report describing scope, affected users, errors, corrective measures and remaining uncertainty without revealing investigation-sensitive data.

Outcome

In this illustrative outcome, Meridian removes the fraudulent network, restores most legitimate educational listings after appeal, compensates affected buyers and reduces recurrence. An independent review finds that detection improved, but language coverage remains weak. The board therefore funds multilingual review and establishes quarterly reporting on health-claim enforcement.

Learning points

The case demonstrates that platform responsibility depends on the whole system, not only on individual posts. Commercial incentives, ranking, seller controls, notice handling, consumer promises and appeal design all affect risk. It also shows why proportionate intervention is superior to indiscriminate blocking: high-risk transactions can be contained while lawful public-health education remains available.

12. Contemporary developments and strategic challenges

Generative AI increases the volume and plausibility of synthetic text, images, audio and video. Risks include impersonation, fabricated evidence, automated harassment, fraud and non-consensual sexual imagery. Governance responses include provenance metadata, watermarking where technically reliable, synthetic-media labels, limits on impersonation, model-provider reporting channels and rapid victim remedies. No single technical signal is universally reliable; labels can be stripped and provenance systems may not cover legacy content.

Cross-border enforcement remains difficult. A removal order issued in one country may conflict with another country's speech protections. Platforms should scope geo-blocking carefully, verify authority, publish government-request statistics and challenge overbroad demands where appropriate. Competition and interoperability concerns also matter: dominant platforms may control both distribution and the evidence needed to assess their conduct.

The central strategic challenge is institutional legitimacy. Platforms should not become unaccountable private governments, but neither should they treat content governance as a purely optional customer-service function. Legitimacy is strengthened by clear rules, proportionality, transparency, accessible remedies, independent oversight and demonstrable attention to affected communities.

Best-practice compliance checklist

A platform compliance programme should:

1. Map every service, content flow, jurisdiction and legal category.
2. Classify the organisation's role and identify applicable safe harbours.
3. Maintain clear rules for illegal, harmful and restricted content.
4. Use risk-based prevention, detection, review and escalation.
5. Keep notice, decision, order and appeal records securely.
6. Provide reasons and meaningful remedies to users.
7. Test automated systems for accuracy, bias, language coverage and drift.
8. Build child safety and accessibility into product design.
9. Protect privacy through minimisation, security and lawful disclosure.
10. Report meaningful metrics and obtain independent assurance for high-risk systems.
11. Train reviewers, managers and engineers in legal and human-rights requirements.
12. Reassess the programme after incidents, regulatory change and major product launches.

Chapter Summary

Intermediary liability is a problem of balancing platform power, public safety, innovation and rights. Safe harbours generally protect qualifying intermediaries from specified third-party liability, but they are conditional and do not cover every claim. Contemporary regulation increasingly combines liability rules with procedural duties, systemic-risk management, transparency, audits and user remedies.

Effective content governance distinguishes illegal content from lawful but harmful material, evaluates context and risk, and uses proportionate interventions. Moderation requires prevention, detection, human judgment, enforcement, appeals and oversight. Misinformation policy must address accuracy, intent, virality and freedom of expression without granting platforms or governments unlimited authority. Child protection, accessibility and privacy are design requirements rather than afterthoughts.

For postgraduate practitioners, the key lesson is systemic: legal accountability depends not only on what a user posts, but also on platform design, ranking, monetisation, notice handling, staffing, evidence, transparency and institutional governance. A defensible programme is rights-aware, risk-based, documented, measurable and adaptable to changing law and technology.

Key Terms

Intermediary: A service that facilitates communication, access, hosting or transactions between third parties.

Platform: A socio-technical service combining infrastructure, interfaces, rules and mechanisms for interaction or distribution.

Safe harbour: Conditional legal protection from specified liability for third-party content or conduct.

User-generated content: Material created or uploaded by service users.

Publisher: An entity that exercises editorial responsibility for content before or during publication.

Content moderation: Processes used to detect, evaluate, restrict or remove content.

Content governance: The wider system of policies, design, operations, accountability and oversight governing content.

Illegal content: Content that violates an applicable legal rule.

Harmful content: Content capable of causing significant harm even when not universally unlawful.

Misinformation: Inaccurate information shared without demonstrated deceptive intent.

Disinformation: Deliberately deceptive or manipulative information.

Malinformation: Genuine information used or disclosed in a harmful context.

Notice-and-action: A process through which a service receives a sufficiently specific report and takes an appropriate response.

Due diligence: Reasonable organisational, technical and procedural measures used to identify and reduce legal and social risks.

Dark pattern: An interface design that manipulates or undermines a user's free and informed choice.

Age assurance: Methods used to estimate or verify whether a user meets an age threshold.

Provenance: Information about the origin, history or authenticity of digital content.

Over-removal: Restricting lawful material because of error, uncertainty or excessive enforcement.

Appeal: A process allowing a user or complainant to request reconsideration of a content decision.

Systemic risk: A risk arising from the design, operation or scale of a platform and affecting society or significant groups.

Review Questions

Short Answer Questions

1. What is intermediary liability, and how does it differ from ordinary publisher liability?
2. Why is safe harbour generally described as conditional rather than absolute?
3. Distinguish illegal content, harmful content, misinformation and disinformation.
4. Identify four procedural safeguards that should accompany a content-removal decision.
5. Why should accessibility be treated as part of content governance?

Long Answer Questions

1. Compare the policy logic of United States Section 230, the European Union Digital Services Act and India's intermediary framework.
2. Explain how a risk-based moderation system can reduce both false positives and false negatives.
3. Analyse the relationship between freedom of expression, privacy and platform enforcement.
4. Discuss the legal and operational challenges of protecting children online without creating disproportionate surveillance.
5. Evaluate why recommendation and monetisation systems are relevant to platform accountability even when a platform did not create the underlying content.

Analytical / Application Questions

1. A platform receives a government request to remove a journalist's post alleging corruption. The request cites national security but provides no specific legal provision or evidence. Design a legally cautious decision process, including authentication, scope, user notice, escalation and appeal.
2. An automated model removes posts in a minority language at twice the rate observed in English, while appeal reversals are substantially higher. Diagnose possible causes and propose a remediation and assurance plan.
3. A video-sharing service leaves a dangerous health rumour online but stops recommending it and adds a warning. Assess whether this is proportionate, identifying facts that would be needed before reaching a final conclusion.

Multiple Choice Questions with Answers

Question 1

What is the central function of an intermediary safe harbour?

A. To guarantee that all user content is lawful

B. To provide conditional protection from specified liability for third-party content

C. To require prior human review of every upload

D. To eliminate all platform duties toward users

Correct Answer: B. To provide conditional protection from specified liability for third-party content

Question 2

Which response best illustrates a proportionate intervention for uncertain but potentially harmful misinformation?

A. Permanent deletion of every post containing a related keyword

B. No action in all circumstances

C. Contextual labelling and reduced recommendation while investigating the claim

D. Automatic suspension of every user who shares the claim

Correct Answer: C. Contextual labelling and reduced recommendation while investigating the claim

Question 3

Which item is most directly a procedural safeguard in content moderation?

A. Opaque ranking criteria

B. Notice and an opportunity to appeal

C. Unlimited retention of identity documents

D. Removal without recording the reason

Correct Answer: B. Notice and an opportunity to appeal

Question 4

Why can a platform's recommendation system matter to accountability?

A. It may amplify reach and contribute to foreseeable risk

B. It converts every user into an employee

C. It makes all content automatically illegal

D. It prevents users from seeing any content

Correct Answer: A. It may amplify reach and contribute to foreseeable risk

Question 5

Which statement correctly distinguishes misinformation from disinformation?

A. Misinformation is always criminal; disinformation is always lawful

B. Misinformation is inaccurate information without demonstrated deceptive intent, while disinformation involves deliberate deception

C. Misinformation concerns images only; disinformation concerns text only

D. There is no meaningful distinction

Correct Answer: B. Misinformation is inaccurate information without demonstrated deceptive intent, while disinformation involves deliberate deception

Question 6

Which measure is most consistent with privacy-preserving child protection?

A. Collecting every user's government identity document by default

B. Using risk-proportionate age assurance and minimising retained data

C. Publishing children's profiles for parental verification

D. Disabling all private communication for adults and children

Correct Answer: B. Using risk-proportionate age assurance and minimising retained data

Question 7

What is a principal danger of relying on a single aggregate moderation-accuracy score?

A. It may conceal poor performance for particular languages or groups

B. It guarantees excessive transparency

C. It prevents any automated moderation

D. It proves that all decisions are legally correct

Correct Answer: A. It may conceal poor performance for particular languages or groups

Question 8

In the illustrative Meridian case, why was a broad keyword sweep inadequate?

A. Keywords can never be used in moderation

B. It treated terms as conclusive and caused over-removal of lawful material

C. It required too much human contextual review

D. It made counterfeit goods legal

Correct Answer: B. It treated terms as conclusive and caused over-removal of lawful material

Question 9

Which governance practice most directly supports external accountability?

A. Destroying appeal records after each decision

B. Publishing meaningful enforcement information and enabling independent audits

C. Keeping all policies secret

D. Allowing only advertisers to review moderation

Correct Answer: B. Publishing meaningful enforcement information and enabling independent audits

Question 10

What is the best description of systemic risk?

A. A risk limited to one user's forgotten password

B. A risk arising from platform design, operation or scale and affecting society or significant groups

C. A risk that exists only in physical retail stores

D. A risk eliminated whenever content is labelled

Correct Answer: B. A risk arising from platform design, operation or scale and affecting society or significant groups