Tuteliq AB's confidential, held-out coded-term / online-slang detection benchmark, companion to the public report. The card is public; the data is gated. Access is request-only, requires a use-case review, and is granted solely at Tuteliq's discretion, governed by the Coded-Terms / Online-Slang Dataset Access Agreement v1.0 (reproduced in full on this card). CSAM-trading codes are excluded and never distributed.

Access is governed by the Coded-Terms / Online-Slang Dataset Access Agreement v1.0 (Tuteliq AB, org. no. 559481-1217), reproduced in full in the dataset card below. By requesting access you, on behalf of yourself and your organization, agree to be bound by that Agreement, and in particular:

CSAM guardrail - CSAM-trading codes are excluded and never distributed; you will never request, reconstruct, or seek them, and will report any you find to legal@tuteliq.ai (s.2).
High-confidence terms only - you will not speculate about or research excluded draft/low-confidence terms (s.2.2).
Permitted use only - research, linguistic analysis, and benchmarking of child-safety detection, approved in writing after use-case review (s.3, s.5).
No production, no training - you will NOT deploy the terms in any production/filtering system, train or fine-tune any model on them, or build commercial products from them (s.4.1).
No redistribution - no sharing outside named Authorized Users; no re-upload to any public repository (s.4.2).
Security - AES-256 at rest, named Authorized Users only, MFA + full-disk encryption, no unencrypted cloud sync, access logs (s.6).
Breach - report within 24 hours to legal@tuteliq.ai (s.7).
Attribution - cite the dataset and the companion public report as required (s.8).
Term - 12 months, no automatic renewal; delete all copies on termination (s.10-s.11).
Governing law - Swedish law; Swedish courts (s.13).

Because of the nature of the content, access requires a detailed use-case review (s.5). Approved recipients receive gated access to this Hugging Face repository, bound to the approved account and Authorized Users list, and must execute the full Agreement. Access is granted solely at Tuteliq AB's discretion. Providing false information voids any grant.

Kids' Online Slang & Coded-Terms Detection Benchmark

The companion evaluation dataset to the Tuteliq research report The State of Kids' Online Slang 2026. It measures whether a detection system recognises coded language — slang, acronyms, and emoji codes children use to discuss risk — in context, and whether it avoids over-flagging benign look-alikes.

Authors: Dr. Nicola Harding, Dr. Gabriel Sabadin — expert review by Prof. Sarah Kingston (University of Central Lancashire)
Maintainer: Tuteliq AB
Version: v1.0 (git tag v1.0)
Size: 315 high-confidence test cases across 14 risk categories + benign controls
Languages: English + Spanish, Portuguese, French, German, Swedish

🔒 Responsible scope — CSAM-trading codes are excluded

This dataset deliberately excludes CSAM-trading codes. Those are handled only through law-enforcement / IWF / NCMEC channels and are never published here or anywhere else. Every included item is high-confidence and expert-reviewed.

⚠️ Held-out — do NOT train on this data

This set exists to measure detectors. Training on it, directly or via leakage, invalidates every result reported against it. Keep it out of all training and fine-tuning corpora.

Motivation

Keyword and single-message classifiers miss coded language: the same token can be a harmful code or an innocent word depending on context ("molly" the drug vs. a friend named Molly; a peach emoji as flirtation vs. fruit). This benchmark rewards context-aware detection and penalises naive keyword matching, using paired harmful cases and benign look-alikes.

Risk categories (14 + benign controls)

sexual_content, sexualization, photo_request, secrecy_request, meeting_request, sextortion, image_based_abuse, grooming, self_harm, eating_disorder, substance, extremism, cyberbullying, fraud, and benign_lookalike (controls that resemble coded terms but are innocent).

Format

testcases.jsonl — one JSON object per line:

{"text": "...", "child_age": 14, "should_flag": true, "category": "substance", "lang": "en", "source_confidence": "high"}

should_flag is the gold label; benign_lookalike rows are should_flag: false and measure precision.

Scoring

Per-case accuracy and F1 against should_flag.
Category-level recall on the harmful classes.
Benign-lookalike false-positive rate (precision of targeting).

Provenance, ethics & safety

Expert-curated, high-confidence coded-term test cases; low/medium-confidence and draft terms are excluded. No real PII, no user content, no CSAM-trading codes. Example texts are constructed to depict coded usage at an indicator level for the sole purpose of evaluating protective detection systems.

Access

The card is public; the data is gated with manual approval. Request access via the button above. Every request is reviewed by Tuteliq AB; no data is downloadable without an approved grant.

Citation

@misc{tuteliq_slang_benchmark_2026,
  title  = {Kids' Online Slang & Coded-Terms Detection Benchmark},
  author = {Harding, Nicola and Sabadin, Gabriel},
  year   = {2026},
  note   = {Held-out evaluation dataset, v1.0. Gated. Companion to "The State of Kids' Online Slang 2026" (Tuteliq AB).},
  url    = {https://huggingface.co/datasets/tuteliq/kids-online-slang}
}

Contact

Web: https://tuteliq.ai
Research & dataset-access enquiries: research@tuteliq.ai
Legal / breach notification: legal@tuteliq.ai

Access requests are handled through the gated request flow above; general questions and research collaboration enquiries are welcome at research@tuteliq.ai.

Full Access Agreement (v1.0)

The complete governing agreement. Approved recipients must execute it; the summary at the access gate is not a substitute for these terms.

Coded-Terms-Slang Dataset Access Agreement

Tuteliq AB
Organization Number: 5594811217
Address: Sweden

1. Parties and Definitions

1.1 Parties

Provider: Tuteliq AB ("Tuteliq"), a Swedish company
Recipient: The individual or organization identified in Section 8 ("Requestor")

1.2 Definitions

"Dataset" means the tuteliq/kids-online-slang dataset hosted on Hugging Face (the "Kids' Online Slang & Coded-Terms Detection Benchmark"), containing:

315 high-confidence, expert-reviewed test cases across 14 risk categories plus benign-lookalike controls, measuring context-aware detection of coded language (slang, acronyms, emoji codes) used by children and adolescents online
Per-case fields: text, child age, gold label (should_flag), risk category, language, and source-confidence
Coverage across English, Spanish, Portuguese, French, German, and Swedish
Source attribution: expert linguists (Harding, Sabadin, with expert review by Kingston); companion to Tuteliq's public report "The State of Kids' Online Slang 2026"
Version: 1.0, release date August 5, 2026
Format: Hugging Face dataset (testcases.jsonl, one JSON object per line)

"CSAM-Trading Codes" means coded language, abbreviations, or terminology specifically used to facilitate or conceal the distribution of Child Sexual Abuse Material (CSAM). These are explicitly and permanently excluded from the Dataset.

"Permitted Use" means research, linguistic analysis, or benchmarking of child-safety detection systems, as defined in Section 3.

"Prohibited Use" means any deployment of the Dataset in production systems, model training, or other uses beyond Permitted Use, including but not limited to:

Integrating terms into filtering systems
Training models on Dataset terms
Commercial product development using Dataset content
Unauthorized redistribution

"Institutional Affiliation" means employment at, or official research engagement with, a university, research organization, or established child-safety platform.

"Use-Case Review" means Tuteliq's evaluation of the Requestor's planned application, data security measures, and organizational intent before access is granted.

2. Critical Guardrails

2.1 CSAM-Trading Codes are Excluded and Never Distributed

This is a non-negotiable principle.

❌ CSAM-trading codes do NOT appear in the Dataset
❌ CSAM-trading codes are NEVER shared on public platforms (including Hugging Face)
❌ CSAM-trading codes are LE/IWF/NCMEC channel only
✅ Tuteliq verifies zero CSAM-trading codes before any Dataset release

Requestor acknowledges and agrees:

The Dataset intentionally excludes CSAM-trading codes
This exclusion protects law enforcement channels and prevents code normalization
Requestor will never request, attempt to reconstruct, or seek CSAM codes from Tuteliq
If Requestor discovers CSAM-trading codes in the Dataset, Requestor will immediately notify Tuteliq (legal@tuteliq.ai)

2.2 Confirmed Terms Only

The Dataset contains only high-confidence, expert-reviewed cases (source_confidence: high), drawn from the confirmed terminology underlying Tuteliq's public report "The State of Kids' Online Slang 2026".

❌ Low-confidence or experimental terms are NOT included
❌ Medium-confidence terms under review are NOT included
✅ Only terms verified through multiple sources and expert consensus are included

Requestor agrees to use only confirmed terms and not to speculate about or research excluded draft/low-confidence terms.

3. Permitted Uses

3.1 Research and Analysis

Requestor may use the Dataset ONLY for:

Academic Research

Linguistic analysis of child-safety terminology and linguistic evolution
Comparative study of coded language across platforms and age groups
Publication of findings in peer-reviewed academic venues
University capstone projects or graduate theses

Benchmarking and Detection System Evaluation

Evaluating existing child-safety detection models against the Dataset
Measuring term-detection accuracy and recall of competitor systems
Academic comparison of detection approaches
Publishing detection performance metrics (without exposing the Dataset itself)

Trust & Safety Research

Understanding terminology trends in child-safety contexts
Analyzing linguistic patterns for detection research
Informing safety policies and guidelines development
Regulatory compliance and child-safety audits

3.2 Analysis Methods

Requestor may conduct:

Term frequency analysis and distribution across platforms
Linguistic pattern analysis (morphology, etymology, semantic clusters)
Comparative co-occurrence analysis (which terms appear together)
Temporal trend analysis (if versioned over time)
Platform-specific variation studies

3.3 Publication and Disclosure

Requestor may publish findings as:

Academic papers and research publications (with attribution to Tuteliq)
Internal safety reports and policy briefings
Conference presentations (without reproducing Dataset content)
Regulatory submissions (describing methodology, not exposing raw terms)

Important Restrictions on Publication:

❌ Do NOT republish the full Dataset or large subsets in publications
❌ Do NOT create a public appendix listing all 367 terms
❌ Do NOT redistribute Dataset content to non-Authorized Users
✅ Cite Tuteliq's public report: "The Kids' Coded Language Report" (Harding, Sabadin, Kingston, 2026)
✅ Reference the Hugging Face dataset card for methodology

3.4 Complementary to Public Report

This Dataset is designed to complement Tuteliq's public "Kids' Coded Language" report. Requestor agrees to:

Acknowledge the public report as primary source material
Direct other researchers to the public report for context and authority
Not position private Dataset access as an alternative to the public report
Support the report's visibility and credibility

4. Prohibited Uses

Requestor STRICTLY PROHIBITS:

4.1 Production Deployment and Training

❌ Integrating Dataset terms into production filtering/detection systems
❌ Training machine learning models on Dataset terms
❌ Fine-tuning language models using Dataset vocabulary
❌ Using Dataset terms in rule-based detection systems deployed to end-users
❌ Building commercial products based on Dataset terms
❌ Sublicensing or white-labeling Dataset content

4.2 Redistribution

❌ Sharing the Dataset with researchers not on the Authorized Users list
❌ Uploading the Dataset to public repositories (GitHub, Zenodo, HuggingFace, etc.)
❌ Depositing Dataset in institutional repositories or archives
❌ Selling or commercializing Dataset access
❌ Publishing the full Dataset as a supplementary material file

4.3 CSAM-Related Activity

❌ Attempting to reconstruct, deduce, or request CSAM-trading codes
❌ Comparing Dataset terms against CSAM databases or law enforcement materials
❌ Sharing Dataset with law enforcement (contact LE@tuteliq.ai for LE access instead)
❌ Using Dataset to evade law enforcement detection systems

4.4 Misuse and Harm

❌ Using terms to target, harass, or identify children
❌ Coaching children on coded language to evade detection
❌ Using Dataset to undermine child-safety initiatives
❌ Competitive intelligence or reverse-engineering Tuteliq's safety models

5. Use-Case Review and Approval

5.1 Use-Case Assessment

Unlike TUT-100 (evaluation only), the Coded-Terms dataset requires detailed use-case review due to the nature of the content.

Tuteliq will evaluate:

Legitimacy: Is the stated use genuinely focused on research or benchmarking?
Organization: Does Requestor have institutional credibility and established safety commitment?
Security: Can Requestor demonstrate data protection measures?
Intent: Is there potential for misuse or harm?
Scope: Are Authorized Users limited and vetted?

5.2 Use-Case Submission

Requestor must provide:

Detailed description of the research question or safety problem
Explanation of why this Dataset is necessary (vs. public report alone)
List of anticipated analyses and outputs
Publication/sharing plan for results
Commitment to academic rigor and child-safety ethics
Evidence of institutional approval (if academic) or corporate safety governance

Example acceptable use cases:

"We are researchers at [University] studying linguistic evolution in child-safety terminology. We will analyze term co-occurrence patterns and compare against published detection benchmarks."
"We are a platform's Trust & Safety team evaluating detection accuracy. We will test our model against the Dataset and publish aggregate performance metrics."

Example rejected use cases:

"We want to build a term-filtering system" → (Production deployment prohibited)
"We want to train an NLP model" → (Training prohibited)
"We need to know secret codes kids use" → (Vague intent, potential misuse)

5.3 Approval Timeline

Tuteliq shall:

Acknowledge receipt within 2 business days
Request clarification if needed (respond within 5 business days)
Conduct use-case review (10–15 business days)
Approve, request conditions, or deny in writing
Issue agreement and access credentials if approved

6. Data Handling and Security

6.1 Secure Storage

Requestor agrees to:

Store the Dataset on isolated, access-controlled systems
Encrypt the Dataset at rest using AES-256 or equivalent
Restrict network access (no cloud sync without encryption)
Maintain audit logs of all Dataset access (who, when, what)
Implement role-based access control (principle of least privilege)

6.2 Authorized Users

Requestor shall:

Limit access to named individuals (primary contact + research team)
Remove access immediately when team members leave
Update Authorized Users list within 5 business days of any changes
Ensure all Authorized Users acknowledge the agreement terms

6.3 Workstation Requirements

Dataset access only on:

Workstations with full-disk encryption
Multi-factor authentication for login
Firewalled networks with intrusion detection
Devices not connected to unsecured/public networks during access

6.4 No Public Cloud Storage

❌ Do not sync to Google Drive, Dropbox, OneDrive, iCloud
❌ Do not upload to shared repositories or file-sharing platforms
✅ Store on local encrypted systems or enterprise-managed infrastructure

6.5 Deletion Upon Termination

Upon termination, Requestor shall:

Permanently delete all copies of the Dataset
Remove from all workstations and backups
Provide written certification of deletion within 10 days
Retain confidentiality obligations indefinitely

7. Breach Notification and Investigation

7.1 Incident Reporting

If Requestor becomes aware of any actual or suspected unauthorized access, disclosure, or loss of the Dataset, Requestor shall:

Notify Tuteliq immediately (within 24 hours) at legal@tuteliq.ai
Provide detailed incident description
Include scope (how many records accessed, which users involved)
Outline containment measures already taken

7.2 Cooperation

Requestor shall:

Preserve all evidence related to the breach
Cooperate with Tuteliq's investigation
Provide forensic logs and access records on request
Implement remedial measures within 30 days

7.3 Termination for Breach

Unauthorized access, redistribution, or training using the Dataset results in immediate termination and potential legal action.

8. Intellectual Property and Attribution

8.1 Ownership

Tuteliq retains all intellectual property rights in:

The Dataset and its contents
The 315 high-confidence test cases and their labels
Linguistic annotations, context metadata, and severity classifications
The methodology and source attribution framework

8.2 Required Attribution

Any research, publications, or presentations using the Dataset must cite:

"We evaluated [model/research] using the tuteliq/kids-online-slang dataset, developed by Tuteliq AB (Harding, Sabadin, with expert review by Kingston). This dataset contains 315 high-confidence coded-term test cases for child-safety detection research. Dataset version 1.0 is available on Hugging Face under a gated-access research license."

8.3 Public Report Attribution

Requestor shall acknowledge the foundational public report:

"The analysis builds on Tuteliq's 'Kids' Coded Language Report' (2026), which provides contextual and linguistic authority for the terminology included in this research."

9. Access Request and Registration

9.1 Requestor Information

To request access, Requestor must provide:

Full legal name and title
Organization name and type (academic, platform company, etc.)
Institutional affiliation verification
Primary contact email and phone
Research question or safety problem statement
Detailed use-case description (Section 5.2)
List of Authorized Users (primary contact + research team, typically 2–5 persons)

9.2 Use-Case Submission

Attach:

Brief research proposal (300–500 words)
Explanation of why private Dataset access is needed beyond public report
Analysis plan and expected outputs
Publication strategy (peer-review, internal reports, conference talk)
Evidence of institutional approval or organizational safety commitment
Data security attestation (see Appendix B)

9.3 Approval Process

Tuteliq shall:

Acknowledge receipt within 2 business days
Conduct use-case review (10–15 business days)
Request clarifications if needed
Approve, approve with conditions, or deny in writing

Tuteliq reserves the right to:

Request additional information
Require modifications to research plan
Deny access without detailed reasoning
Revoke access if conditions are violated

9.4 Access Credentials

Upon approval, Requestor receives:

Hugging Face dataset access (tuteliq/kids-online-slang), granted to the approved account
Dataset documentation and methodology guide
Research ethics guidelines and publication best practices
Contact information for ongoing questions

10. Term and Renewal

10.1 Initial Term

This agreement is valid for 12 months from approval date.

10.2 Annual Renewal

Requestor must renew annually by submitting:

Confirmation of continued compliance
Updated Authorized Users list
Summary of research conducted (publications, conferences, findings)
Evidence of continued institutional affiliation
Renewed security attestation
Attestation that no Breaches occurred

Renewal requests must be submitted 60 days before expiration.

10.3 Renewal Decision

Tuteliq shall:

Review renewal requests within 20 business days
Consider research outcomes and publication record
Approve renewal or deny and revoke access

No automatic renewal. Access expires at 12 months unless renewed.

11. Termination

11.1 Termination by Tuteliq

Tuteliq may terminate immediately if:

Requestor materially breaches confidentiality or security terms
Unauthorized redistribution, training, or production deployment is discovered
CSAM-trading codes are accessed, requested, or discussed in relation to the Dataset
Requestor fails to renew on schedule
Authorized Users violate access restrictions
Regulatory or legal obligations require termination
Research outcomes misrepresent the Dataset or harm child safety

11.2 Termination by Requestor

Requestor may terminate with 60 days' written notice to legal@tuteliq.ai.

11.3 Effect of Termination

Upon termination, Requestor shall:

Immediately cease all use of the Dataset
Permanently delete all copies (no recovery)
Provide written deletion certification within 10 days
Cooperate with final compliance audit
Retain confidentiality obligations indefinitely

12. Liability Limitation

12.1 No Warranty

The Dataset is provided "AS-IS" without warranty regarding:

Accuracy, completeness, or representativeness of terms
Suitability for any particular research purpose
Non-infringement of third-party rights
Fitness for commercial use (which is prohibited anyway)

12.2 Liability Cap

Tuteliq's total liability under this agreement shall not exceed:

The amount paid by Requestor for access (if any), OR
USD $0 if access is provided at no cost

Tuteliq is not liable for:

Lost profits, research impact, or publication success
Indirect, incidental, or consequential damages
Third-party claims related to Requestor's use
Regulatory fines or penalties from Requestor's Breach

13. Governing Law and Dispute Resolution

13.1 Governing Law

This agreement is governed by Swedish law.

13.2 Venue

Disputes shall be resolved in Swedish courts.

13.3 Pre-Litigation Resolution

Before formal proceedings:

Exchange written notice of dispute
Attempt good-faith negotiation (30 days)
Escalate to senior management if needed

14. Compliance and Audit

14.1 Audit Rights

Tuteliq reserves the right to:

Request audit documentation of data security measures
Conduct remote or on-site security assessments (10 days' notice)
Verify Authorized Users list and access controls
Review access logs and compliance records

Requestor shall provide requested documentation within 15 business days.

14.2 Regulatory Compliance

Requestor warrants that:

All use complies with applicable laws
Requestor has authority to enter this agreement
Research is conducted in accordance with institutional ethics policies (if applicable)

14.3 Child Safety Commitment

Requestor acknowledges that this Dataset exists to support child-safety research. Requestor agrees:

The Dataset will be used only for legitimate child-safety purposes
Results will not be misused to harm children
Any findings undermining child safety will be reported to Tuteliq
Requestor opposes the use of coded language to conceal exploitation

15. Amendments and Notices

15.1 Amendments

Tuteliq may amend this agreement:

With 30 days' written notice for material changes
Immediately for clarifications or administrative changes
Continued use constitutes acceptance

15.2 Notices

To Tuteliq:
Tuteliq AB
Legal / breach notification: legal@tuteliq.ai
Research & dataset-access enquiries: research@tuteliq.ai
Web: https://tuteliq.ai
Sweden

To Requestor:
[Contact information from Section 9]

16. Entire Agreement and Severability

This agreement supersedes all prior negotiations and constitutes the entire agreement. If any provision is invalid, the remaining provisions remain in effect.

17. Signatures

For Requestor:

Organization Name: _________________________________

Primary Contact Name: _________________________________

Title: _________________________________

Email: _________________________________

Phone: _________________________________

Authorized Users (attach as Appendix A)

I/we certify authority to bind this organization and agree to all terms.

Signature: _________________________________

Date: _________________________________

For Tuteliq AB:

Representative Name: _________________________________

Title: _________________________________

Signature: _________________________________

Date: _________________________________

Appendix A: Authorized Users and Use-Case Summary

Name	Email	Title	Role	Affiliation	Start Date
(Primary Contact)

Research Question / Safety Problem:

Anticipated Analyses:

Publication Plan:

Data Security Measures:

Appendix B: Data Handling and Security Attestation

Requestor confirms the following security and compliance measures:

Data Storage & Access:

☐ Dataset stored on encrypted systems (AES-256 at rest)
☐ Access restricted to named Authorized Users only
☐ Multi-factor authentication enabled for user accounts
☐ Workstations have full-disk encryption
☐ No cloud synchronization or unencrypted cloud storage
☐ Firewalled networks with intrusion detection

Monitoring & Audit:

☐ Audit logs maintained for all Dataset access (who, when, what)
☐ Logs retained for minimum 12 months and provided upon request
☐ Regular security reviews conducted (at least quarterly)
☐ No unapproved external transfers or sharing

Breach & Incident Response:

☐ Breach notification procedures understood (24-hour reporting to legal@tuteliq.ai)
☐ Incident response plan in place
☐ Key contacts identified for emergency escalation

Regulatory & Ethical Compliance:

☐ Research approved by institutional review board (if applicable)
☐ No training or production deployment of Dataset terms
☐ No attempts to access, reconstruct, or request CSAM-trading codes
☐ No redistribution to unauthorized parties
☐ Child-safety ethics acknowledged and endorsed

Commitment to Terms:

☐ Annual renewal deadline (12 months from approval) understood
☐ Permanent deletion upon termination understood
☐ Confidentiality obligations survive termination indefinitely
☐ Authorized Users acknowledge receipt of this agreement

Authorized Users Acknowledgment:

Name	Email	Signature	Date

Appendix C: Public Report Cross-Reference

This Dataset is complementary to and derived from Tuteliq's foundational public research:

"The State of Kids' Online Slang 2026"
Authors: Dr. Nicola Harding, Dr. Gabriel Sabadin
Expert Review: Prof. Sarah Kingston (University of Central Lancashire)
Published: August 5, 2026
Access: https://tuteliq.ai/reports

The public report provides:

Context for the confirmed terminology behind the 315 test cases
Linguistic methodology and validation
Platform distribution data
Severity classifications
Historical trends and evolution

Researchers using the private Dataset should:

Cite the public report as primary authority
Direct readers to the public report for background and methodology
Acknowledge that public report is the authoritative source
Reference Dataset as a supplement for detailed analysis

Document Version: 1.0
Effective Date: August 5, 2026
Classification: Confidential — Tuteliq Use

Downloads last month: 17

Total file size:

121 kB