Afghan Language Content Moderation — Pashto and Dari Reviewers for DSA-Scoped Platforms

Ariana Nexus moderates Afghan language content for DSA-scoped platforms — native-speaker review, classifier evaluation, and systemic-risk assessment across 24 Afghan languages, Pashto and Dari first. Reviewers read the language and the context, so harmful content is judged by people who recognize it rather than inferred by a classifier that never learned it.
Reviewed at scale. Unread in 24 languages.
Covered by current classifiersAfghan-language · unreviewed
24 languages in scope — 0 with native-grade classifiers.
Harmful content in Pashto, Dari and 22 more moves through DSA-scoped platforms faster than any monolingual stack can see it. Illustrative.
20.5M
CyberTipline reports to NCMEC in 2024 — generative-AI reports up 1,325% year over year
~20M/day
content-moderation decisions logged in the EU DSA Transparency Database
88/100
LLM responses that became health disinformation on instruction — Annals of Internal Medicine, 2025
9%
of adults are confident they can identify a deepfake — Ofcom, 2024

What is Afghan language content moderation?

Afghan language content moderation is the review and classification of user content in Pashto, Dari, and the other 22 Afghan languages by people who speak them. It covers detection of harmful content, evaluation of the classifiers that miss it, and the systemic-risk evidence a regulator expects — work a monolingual moderation stack cannot produce by translation alone.

Ariana Nexus is a Washington, D.C.–area firm providing Afghan language services and cultural intelligence — interpretation, translation, cultural training, compliance support, and AI data — across 24 Afghan languages.

The harm your platform answers for is in languages your systems cannot read

The harmful content your platform answers for is in languages your safety systems never learned to read.

A platform's trust and safety stack — its classifiers, its moderators, its escalation paths — works in the languages it was built for. The terrorism recruitment, the abuse, the gender-based violence, and the targeted harm in Pashto and Dari run underneath it, undetected and unreviewed, because nothing in the pipeline reads the language.

The law does not accept invisibility as a defense. Under the DSA, very large platforms and search engines must assess and mitigate systemic risks every year, and the regulation states that the assessment must take regional and linguistic aspects into account. The UK Online Safety Act imposes parallel duties under Ofcom. A risk nobody can read is a risk nobody assessed.

The systems meant to close the gap are themselves an attack surface. The AI classifiers doing the moderation can be evaded, manipulated, and jailbroken — most easily in the low-resource languages they were trained on least.

Ariana Nexus is the layer that reads what the stack cannot: native-speaker detection and classification across all 24 Afghan languages, classifier evaluation and security testing, and systemic-risk assessment and mitigation documented for the DSA and the Online Safety Act — speech-protective, reviewer-protected, and strictly within legal frameworks.

DSA Art 34(2)
— the law requires accounting for regional and linguistic aspects; most assessments do not.
24
— Afghan languages your moderation stack likely does not cover.
The Multilingual Trust & Safety Standard
— the Ariana Nexus coverage method.
40languages a typical moderation stack covers
+24Afghan languages it likely does not
The coverage gap — 40 languages a typical stack covers; 24 Afghan languages remain uncovered (gold outline): the blind spot a DSA systemic-risk assessment must account for.
Visualization: of sixty-four language tiles, forty light up as covered by a typical moderation stack; twenty-four Afghan-language tiles remain dark with a gold outline, representing the coverage gap.
OPERATING MODEL

One practice. Three coordinated capabilities.

HIC · HUMAN INTELLIGENCE COLLECTIVE

The people who read what classifiers cannot

Lived-expertise practitioners across all 24 Afghan languages; the cultural gatekeepers who keep every engagement anchored in ground truth, never extractive.

Afghan native-speaker reviewers across all 24 languages who read, classify, and contextualize harmful content, catching the coded harm classifiers miss — with reviewer wellbeing protected throughout.

Protocol: The Multilingual Trust & Safety Standard.
ADF · AI DATA FACTORY

The data and the security testing

Governed Afghan-language data infrastructure, evaluation benchmarks, and institutional-grade training assets meeting auditable standards.

Classifier-evaluation data and harmful-content taxonomies mapped to the DSA risk categories; in-language test and red-team sets; OWASP-LLM-aligned security testing of moderation AI; de-identified analytics.

Protocol: The ADF Pipeline.
CCB · CULTURAL COMPLIANCE BUREAU

The governance layer

An audit-grade review regime translating cultural intelligence into compliance-ready practice — the governance layer threading through every engagement.

Risk methodology mapped to DSA Articles 34/35 and the Online Safety Act; cultural and contextual validation of classifications; reviewer-wellbeing, rights, and speech-protection governance; the CCB Sign-Off Mark on risk assessments.

Protocol: The CCB Sign-Off Mark.

How Ariana Nexus closes the coverage gap: the Multilingual Trust & Safety Standard

The Multilingual Trust & Safety Standard closes the language-coverage gap in content moderation; the Five-Gate Validation Protocol governs the rigor, the rights, and the reviewer.

The Five Gates

GATE 1

Linguistic Accuracy

Harmful-content detection and classification linguistically accurate across all 24 languages and dialects.

GATE 2

Cultural Validity

Contextual and cultural validity of classifications, since harm is context-dependent and often coded; the Cultural Hallucination Audit applied to moderation classifiers; cleared by the CCB Sign-Off Mark.

GATE 3

Standards Conformance

DSA Articles 34 and 35, the UK Online Safety Act, and the OWASP Top 10 for LLM Applications; DSA transparency and independent-audit obligations.

GATE 4

Population Risk

Speech-protective and viewpoint-neutral, targeting illegal and harmful content rather than lawful expression; reviewer wellbeing protected; privacy and civil liberties preserved; no handling of illegal content, which remains with legally authorized entities.

GATE 5

Institutional Sign-Off

Risk assessments, mitigation measures, and classifier evaluations documented, traceable, and ready for DSA and Online Safety Act regulators and independent auditors.

The Four-Phase Delivery Cycle

Diagram: the four-phase delivery cycle traced as a continuous path from Phase one Situation through Phase four Measured Outcome.

I · Situation — Understand.

The platform or system, its Afghan-language exposure, and its DSA and Online Safety Act obligations mapped; the in-language systemic-risk picture assessed.

Cultural mapping · stakeholder calibration · constraint discovery.

II · Complication — Architect.

The multilingual coverage plan, classifier-evaluation framework, DSA-aligned harmful-content taxonomies, and mitigation measures designed.

Program scaffolding · compliance baseline · governance charter.

III · Resolution — Deploy.

In-language detection and classification evaluated and improved; mitigation implemented; harmful content addressed within legal frameworks; reviewers supported.

In-context execution · data infrastructure.

IV · Measured Outcome — Govern.

Systemic-risk assessment and mitigation documented for the DSA and Online Safety Act; classifier performance monitored; coverage maintained as content and threats evolve.

Continuous documentation · red-team validation · multi-decade horizon.
All three capabilities run throughout: harm in low-resource languages is where human judgment, governed data, and cultural validation each matter most.
DSA ARTICLES 34 & 35

The systemic-risk categories the law names — mapped to the languages it requires you to account for

Illegal contentFundamental-rights harmsCivic & electoral integrityGender-based violenceHarms to minorsCoverage across24 Afghan languagesASSESSED · MITIGATED · DOCUMENTED
Illegal content→ coverage across 24 languages
Fundamental-rights harms→ coverage across 24 languages
Civic & electoral integrity→ coverage across 24 languages
Gender-based violence→ coverage across 24 languages
Harms to minors→ coverage across 24 languages

The obligations you answer to — with their status, as of 2026

The instruments that govern trust, safety, and content integrity for a healthcare-relevant platform. Status is tracked, not assumed: a proposal is not a law, and a contested rule is flagged as such.

Instrument
Scope for trust & safety
Status
EU Digital Services Act
Reg. (EU) 2022/2065 · Arts. 34–35, 37
Annual systemic-risk assessment, mitigation, and independent audit for very large platforms. Public health is a named systemic-risk category. First fine — €120M against X — adopted Dec 2025, under appeal.
In force
Applicable 17 Feb 2024
EU AI Act
Reg. (EU) 2024/1689 · Art. 50, Ch. V, Annex III
Labeling of AI-generated and manipulated content, AI-interaction disclosure, and high-risk duties. Health chatbots fall under Art. 50; clinical AI is high-risk.
Phased
Art. 50 + high-risk: 2 Aug 2026
UK Online Safety Act 2023
c. 50 · Ofcom Illegal Harms & Children Codes
Duty to assess and mitigate illegal content and protect children, with highly effective age assurance. Suicide, self-harm, and eating-disorder content is Primary Priority Content.
In force
Children's duties from 25 Jul 2025
US TAKE IT DOWN Act
Pub. L. 119-12 · FTC enforcement
Criminalizes non-consensual intimate imagery, including AI deepfakes, and requires platforms to remove reported material within 48 hours.
In force
Platform duty live 19 May 2026
Kids Online Safety Act
S. 1748 (119th Congress)
Would impose a minor-protection duty of care and safe-by-default design. It is not law: it died in the House in 2024 and is pending in committee.
Proposed
No floor vote
Section 1557 — Nondiscrimination
45 CFR 92.210 · 89 Fed. Reg. 37522
Bars discrimination through clinical decision-support tools and AI in patient communications. The decision-support provision took effect, but portions are vacated or enjoined.
Litigated
DSI duty effective 1 May 2025
ONC / ASTP HTI-1 Final Rule
45 CFR 170.315(b)(11)
Source-attribute transparency for predictive and AI decision-support in certified health IT — thirty-one disclosed attributes for predictive interventions.
In force
Compliance 31 Dec 2024
C2PA / Content Credentials
C2PA spec v2.x · Content Authenticity
Cryptographically signed media provenance to counter synthetic and misleading content. An open standard with broad adoption — no legal mandate.
Voluntary
Conformance program, 2025
NIST AI RMF + ISO/IEC 42001
NIST AI-600-1 · ISO/IEC 42001:2023
Voluntary frameworks for managing AI risk — including information integrity and confabulation — and a certifiable AI management system.
Voluntary
GenAI Profile, Jul 2024
ONLINE SAFETY ACT · UKDSA · EUROPEAN UNIONGULF & ARAB STATESUnited StatesUnited KingdomSwedenNetherlandsGermanyFranceAustriaItalyUAEQatarSaudi Arabia
United States · anchor
United Kingdom · Online Safety Act (Ofcom)
European Union · DSA Articles 34/35
Gulf & Arab states with significant Afghan communities
Status as of 14 June 2026. Proposals are marked as such and carry no legal force until adopted; contested rules are flagged “litigated.” Not legal advice — confirm against the current text before relying on any item.
THE RECEIVABLES

Your safety systems, extended to the languages they miss

The bench behind the coverage: Afghan native-speaker reviewers with the dialect range, subject knowledge, and institutional standing the work requires.

What an engagement returns

A multilingual trust-and-safety coverage assessment.

Where your systemic-risk obligations meet your language blind spots — DSA Article 34/35 aligned.

In-language harmful-content detection and classification evaluation.

How your classifiers actually perform in Pashto, Dari, and 22 more.

A Cultural Hallucination Audit of your moderation models.

The fluent-but-wrong classifications that pass automated review.

OWASP-LLM-Top-10-aligned security review of your moderation AI.

The model doing the moderation, tested as an attack surface.

Systemic-risk assessment and mitigation documentation.

Ready for the DSA, the Online Safety Act, and independent audit.

Native-speaker human review, with reviewer wellbeing protected.

Real judgment, humanely sourced.

Within legal frameworks, always.

The firm provides expertise and evaluation; it does not handle illegal content, which stays with authorized entities.

Speech-protective and rights-respecting throughout.

Harmful content addressed; lawful expression protected.

Questions platform teams ask before they buy

Which Afghan languages does coverage include?

Ariana Nexus covers all 24 Afghan languages, with a standing bench in Pashto and Dari and additional Afghan languages sourced on defined notice. Pashto content moderation and Dari review run continuously; Uzbeki, Turkmeni, Balochi and the rest are staffed to the volume a platform actually sees. Coverage is stated per language, never averaged.

Can a Farsi-speaking reviewer moderate Dari content?

Not reliably. Afghanistan Dari and Iranian Persian differ in vocabulary, register, and idiom, and the slang that carries harm is exactly where they diverge. A Farsi reviewer will pass content an Afghan reviewer flags, and flag content that is harmless. Ariana Nexus staffs Afghan reviewers — Pashto content reviewers and trust and safety Dari reviewers — and records the dialect on every decision.

Why can't machine translation plus an English classifier cover Pashto?

Because harm survives translation badly. Machine translation flattens the euphemism, the honorific, and the coded reference that make a Pashto post threatening, then hands a clean sentence to a classifier trained on English. Harmful content detection in Pashto depends on that context, so Ariana Nexus pairs native-speaker review with Afghan language classifier evaluation and measures the gap rather than assuming it.

Is Hazaragi a separate language for moderation?

Hazaragi is a dialect of Dari, not a separate language, though platforms often log it separately. It carries distinct vocabulary, and a reviewer who knows only standard Dari can miss both the insult and the threat. Ariana Nexus treats Hazaragi as a Dari variety and matches reviewers to the variety the content is actually in.

How much does Afghan language content moderation cost?

Cost follows volume, language mix, turnaround, and whether the work is queue review, classifier evaluation, or a documented systemic-risk assessment. A one-off evaluation set is a different order of work from standing queue coverage. Ariana Nexus scopes against actual content volume before quoting, and publishes no rate card, because a rate quoted without volume is a guess.

Do DSA obligations apply to Afghan-language content?

Yes, where the service reaches EU users. The Digital Services Act requires very large platforms to assess and mitigate systemic risks across their service, and it does not exempt languages the platform chose not to staff. That makes DSA systemic risk in Afghan languages a documentation problem as much as a staffing one, and Ariana Nexus produces the evidence in both forms.

THE DOOR

Request a Trust & Safety Coverage Review.

For platforms, including very large online platforms and search engines; AI developers building moderation systems; trust-and-safety teams; and content-integrity functions. Speech-protective and within legal frameworks. Briefings are conducted under NDA, in Washington, D.C. or virtually.

Request a confidential briefingHave a specific scenario you would like assessed — a language, a content category, a regulatory deadline? Bring it to the briefing.