←Deliverables
14 placeholder(s) for the team to fill
Report · Parts A to H with three appendices

Sukoon: Professional Practices Compliance and Ethics Audit

CS4001 Professional Practices in IT · BSCY-PPIT Fall 2026 · Submitted 20 November 2026 FAST-NUCES Islamabad · Group: Muhammad Faizan Shakeel (23i-2074) · Muhammad Hayyan (23i-2041) · Abdul Ahad (23i-2014) Repository and live demo: github.com/MuhammadFaizan300/Sukoon-PPIT

Executive Summary

Sukoon is a humanitarian technology product: an AI-assisted mental-health first-aid app for young people in Pakistan. The brief asked us to act as compliance officers and ethical auditors rather than founders, so we designed the product and then tested every feature against five professional-practice filters (ethics, data privacy, cyber crime, intellectual property and organisation) before accepting it.

The central question was whether the product's features, revenue model or operations could violate data privacy law (GDPR and the PDPB), infringe intellectual property, or trigger offences under PECA 2016 such as data theft, dark-web exposure or attention-economy exploitation. In summary:

Alongside this report, the repository contains a working prototype (the bonus requirement), twelve interactive simulations and a 15-slide pitch deck that embeds them. Everything runs on synthetic data, and Part H states exactly what is simulated. The sample Privacy Policy and Terms of Service, the industry validation record and the individual reflections are in Appendices 1 to 3.

Part A: Product Vision and Technical Planning

A.1 The idea

Problem. Many young people in Pakistan live with anxiety, depression or crisis but have little access to specialists, and stigma pushes them to hide it. They will only take a first step online if they trust that it will not expose them.

Solution. Sukoon offers anonymous first-aid support in Urdu, Roman Urdu and English, a private journal, access to verified counsellors and a safe route to human help in a crisis. It is a service to humanity: revenue exists only to keep it running.

A.2 Core features

#FeatureWhat it doesCompliance decision built in
1Anonymous AI support chatSupportive conversation and crisis detectionConsent-gated; PII redacted before the model; 30-day retention
2Encrypted mood journalPrivate entries the user can rereadAES-256-GCM with a key per user; staff cannot read entries
3Verified counsellor bookingConnects users to licensed counsellorsLeast-privilege access; the user must consent to sharing
4Crisis escalation and safety planRoutes imminent risk to a human and to emergency resourcesAudited break-glass access; no covert tracking
5Privacy Control CenterOne screen for consent, export and erasureConsent can be withdrawn in one tap; erasure destroys the key

A.3 Technical model and data

Architecture. A mobile-first client talks over TLS to an API gateway in a Pakistan cloud region. Every request passes through rate limiting and role-based access control. Chat text goes through a PII-redaction service before it reaches the crisis classifier. Stored data sits in an encrypted database whose keys are held in a separate key-management service (KMS) inside Pakistan, and a tamper-evident audit log records every sensitive action. The interactive Data-Flow X-Ray in the demo shows each step.

AI model and data sources. The production design uses a multilingual transformer (an XLM-R class model) fine-tuned for crisis detection. It is trained only on licensed clinical text corpora and on synthetic examples written by the team. It does not scrape social media or forums, which would raise copyright, platform-terms and consent problems, and it never uses dark-web or leaked datasets: receiving unlawfully obtained data may itself be an offence under sections 3 and 4 of PECA 2016. User messages are never used for retraining unless the user opts in separately.

What is simulated. No model was trained for this project. The prototype uses a transparent phrase lexicon in place of the model, so that every decision is explainable and the bias audit can be reproduced. Part H records this.

Part B: Data Privacy and GDPR and PDPB Compliance

B.1 Scope and approach

GDPR applies to Sukoon for users located in the EU or EEA (Art. 3(2)), for example members of the Pakistani diaspora. For users in Pakistan the relevant instrument is the Personal Data Protection Bill (PDPB). The bill has been circulated in several drafts and, as of November 2026, we found no evidence that a comprehensive data protection law has been enacted; the 2023 bill was withdrawn in the Senate. We therefore treat the PDPB as a draft. Because both frameworks share the same core principles, we use GDPR as the design benchmark for every user, which also keeps Sukoon compliant if the PDPB is enacted.

B.2 Data inventory

DataExampleSensitivityLawful basisRetentionLocation
Pseudonymous profile"Moon-17", languageLowContractUntil erasurePakistan
Chat text (redacted)"I feel hopeless"Special category (health)Explicit consent, Art. 9(2)(a)30 daysPakistan, encrypted
Journal entriesFree textSpecial categoryExplicit consent365 days or until erasurePakistan, encrypted
Consent receiptsPurpose, time, versionLowLegal obligationUntil erasurePakistan
Audit logActor, action, identifierLow (no content)Legitimate interest, accountabilityRetained with pseudonymous identifiersPakistan

B.3 Privacy audit against GDPR and the PDPB

RequirementReferenceStatusHow Sukoon complies (evidence in code)
ConsentGDPR Art. 6, 7 and 9; PDPB consentCompliantOne purpose per switch, all off by default, a receipt for every change, immediate withdrawal. consent.ts, 9 tests
Data minimisationArt. 5(1)(c)CompliantNames, phone numbers, CNIC numbers, cities, emails and links are redacted before the AI sees the text, and only redacted text is stored. redact.ts, 7 tests
Storage limitationArt. 5(1)(e)CompliantChats are purged after 30 days and journals after 365. purgeExpired() test
Right to be forgottenArt. 17CompliantRows are deleted and the user's key is destroyed, so even backups become unreadable (crypto-shredding). erase() tests
Access and portabilityArt. 15 and 20CompliantOne-tap JSON export of all the user's data. exportData() test
Security of processingArt. 32CompliantPer-user AES-256-GCM, role-based access control and a hash-chained audit log. crypto.ts, audit.ts
Data protection by designArt. 25CompliantThe consent gate and encryption are defaults, not options
AccountabilityArt. 5(2) and 30CompliantTamper-evident audit log and records of processing
Special category dataArt. 9CompliantExplicit consent for each purpose; no secondary use
ChildrenArt. 8PartialMinimum age of 16, or parental consent; age assurance is planned future work

B.4 Data protection features

Protection against cybercriminals is layered. Encryption protects data in transit (TLS 1.3) and at rest (AES-256-GCM with a fresh IV for every field, a key per user and tamper detection through GCM authentication). Access control limits counsellors to assigned users who have given consent, reserves break-glass access for audited emergencies and requires MFA for staff. Secure storage keeps keys in a separate KMS and encrypts backups. Monitoring relies on the hash-chained audit log and rate limiting. The test suite shows that tampered ciphertext is rejected, that one user's key cannot read another user's data, and that a destroyed key makes data unrecoverable.

B.5 Cross-border data flow

RouteDataVerdictRequired safeguards
Pakistan to PakistanAnyAllowedNone; this is the default
Pakistan to the EUSensitiveConditionalExplicit consent, standard contractual clauses (SCCs) and keys held in Pakistan
Pakistan to the USSensitiveConditionalExplicit consent, keys in Pakistan, and SCCs or the Data Privacy Framework
Pakistan to ChinaSensitiveBlockedNo adequate safeguards for health data
EU to PakistanSensitiveConditionalSCCs and a transfer impact assessment (TIA), as there is no EU adequacy decision for Pakistan
AnyAnonymisedAllowedTruly anonymised data falls outside the scope of both laws

Storing data outside Pakistan without these safeguards would breach the draft PDPB transfer limits and, for EU residents, Chapter V of GDPR. Sukoon is therefore hosted in Pakistan first, and its keys stay in Pakistan even when any processing happens abroad. The interactive Cross-Border Globe evaluates each route. These rules are an educational model, not legal advice.

B.6 Privacy Policy

The sample Privacy Policy and Terms of Service are in Appendix 1.

Part C: Revenue Model and Attention Economy Audit

C.1 Revenue standard check

ModelRelies on the attention economy?Incentive to harvest dataPECA trigger riskVerdict
AdvertisingYes: revenue grows with time spent in the appHighSection 4, transmission to brokersRejected
Data sellingYesVery highSections 3 and 4, dark-web exposureRejected
Freemium with adsYesHighSection 25 spamming, section 4Rejected
Subscription (counsellor tier)NoLowNoneAdopted
NGO and institutional sponsorshipNoNoneNoneAdopted

Sukoon does not rely on the attention economy. It succeeds when users get help and leave, not when they stay. The Attention-Economy Simulator compares an ad-funded version, which accumulates seven violations in fifteen simulated minutes, with the ethical version, which accumulates none and suggests a break after twenty minutes. All rates in the simulator are illustrative estimates.

C.2 Cyber crime trigger in the revenue model

A model that earns per impression or per data point creates a standing incentive to collect more than is needed, to keep distressed users engaged and to pass data to third parties. Those behaviours map directly to section 4 of PECA 2016 (unauthorized copying or transmission of data) and expose data to resale and dark-web markets. Removing the incentive removes the trigger.

C.3 Feature audit

FeaturePrivacy or cyber crime concernDecision
Infinite scrollingCompulsive use by distressed users; ACM 1.2Banned
Loot box style rewardsGambling mechanics aimed at minorsBanned
Targeted adsBuilt on mood history; GDPR Art. 9Banned
Streak guilt-tripsManipulation of vulnerable usersBanned
Push notificationsSection 25 spammingConsent only, rate limited

C.4 Mitigation and business model

Privacy by design and data minimisation protect users while the product earns ethically. Sukoon would be structured as a social enterprise: a licensed not-for-profit company, for example under section 42 of the Companies Act 2017 (to be confirmed with counsel), funded by counsellor-tier subscriptions, institutional sponsorship and grants. This sustains the service and serves its humanitarian purpose without selling any data.

Part D: Cyber Crime Prevention and PECA 2016

D.1 Cyber crime trigger check

The section numbers and titles below (3, 4, 16, 24 and 25) were checked against the text of PECA 2016. The Act was amended in 2025, so counsel should confirm that no sections have been renumbered.

PECA offenceHow a product like Sukoon could enable itControl
s.3 Unauthorized access to information system or dataStaff or attackers reading user recordsRole-based access control, MFA, consent gates, the audit chain and rate limits
s.4 Unauthorized copying or transmission of dataExporting or selling user dataNo third-party sharing; ciphertext at rest; only the user can export
s.16 Unauthorized use of identity informationAccount takeoverMFA, breached-password checks and account lockout
s.24 Cyber stalkingA counsellor or partner tracking a userNo covert tracking; no contact outside the platform; safety contacts controlled by the user
s.25 SpammingRe-engagement messagingConsent-only, rate-limited notifications

D.2 Dark web and breach risk

Mental-health data is among the most damaging data to leak. In the 2020 Vastaamo case in Finland, psychotherapy records were stolen and used to extort patients, the same pattern as the gaming-platform breach in the brief. Could Sukoon's features expose data in the same way? A naive design could: plaintext rows, injectable endpoints, guessable identifiers and unencrypted backups each lead to a leak. Our Breach Attack Simulator runs the same four scripted attacks (SQL injection, credential stuffing, insecure direct object references and backup theft) against a vulnerable twin and against Sukoon. The twin leaks 25,000 synthetic plaintext records, worth an estimated PKR 3,750,000 on a criminal market. Sukoon blocks three of the attacks and yields only ciphertext from the fourth, which is worth nothing. The simulation is scripted, runs on synthetic data and contains no exploit code.

D.3 Incident response plan

If a breach occurs, the plan runs against a clock that starts at detection.

DeadlineStepOwnerBasis
1 hourContain: isolate systems and revoke keys and sessionsSecurity LeadRunbook
4 hoursAssess the scope, the data categories and the riskDPOGDPR Art. 33(3)
24 hoursReport to the national cybercrime authorityLegal CounselPECA 2016; internal target
72 hoursNotify the data protection authorityDPOGDPR Art. 33
72 hoursNotify affected users in plain language if the risk is highProduct OwnerGDPR Art. 34
7 daysForensic report and root causeSecurity LeadGDPR Art. 33(5)
14 daysBlameless post-mortemEngineering ManagerSecure SDLC

PECA offences are investigated by the National Cyber Crime Investigation Agency (NCCIA), which was set up in 2024 to replace the FIA Cyber Crime Wing and confirmed by the PECA (Amendment) Act 2025. Reporting routes should still be confirmed with counsel. The interactive 72-Hour Incident Clock lets the presenter run this exercise live.

Part E: The Ethical and Professional Code Filter

E.1 Bias and harm assessment

An AI crisis detector can harm the very groups it is meant to protect. We measured recall, the share of genuine crisis messages that the model flags, on 17 synthetic messages written by the team. The English-heavy baseline flagged 100% of English messages but only 17% of Roman Urdu and 20% of Urdu messages, and it caught 40% of male-voiced messages and none of the female-voiced ones. That would breach ACM 1.2 (avoid harm) and ACM 1.4 (be fair and do not discriminate). After the data was rebalanced, every group reached 100%. We therefore require every group to be within 10 percentage points of the best group before any release, and we send remaining misses to a human review queue. These figures illustrate the method on a lexicon stand-in; they are not claims of clinical accuracy.

Dark patterns. Sukoon uses none to collect data. Consent options carry equal visual weight, nothing is pre-ticked, declining never degrades the service, and the app encourages breaks instead of streaks.

E.2 Code of ethics mapping

ClauseHow it applies to Sukoon
IEEE Code, Rule 1: public safety, health and welfareImminent-risk messages escalate to a human and to emergency resources; the safety plan is reviewed by a clinician
ACM 1.1: contribute to society and human well-beingWell-being comes before engagement; no attention-economy features
ACM 1.2: avoid harmCrisis-aware replies, the bias gate and a ban on dark patterns
ACM 1.4: be fair and take action not to discriminateRecall gate for every language and gender
ACM 1.6: respect privacyJournal encryption, minimal data, least-privilege access and user-controlled erasure
ACM 2.9: design robustly and usably secure systemsSecure SDLC, role-based access control and audited break-glass access

The interactive Ethics and Law Constellation links each feature to every clause it satisfies.

E.3 Ethical analysis: the dilemma

Should we sell user data to advertisers? Four ethical lenses give the same answer. Utilitarian: advertising revenue could subsidise free care, but re-identifying even one distressed user causes severe and irreversible harm, so the expected harm outweighs the benefit. Deontological: treating health data as a commodity uses people merely as a means, and consent cannot be bundled or implied. Virtue ethics: a trustworthy practitioner would not be willing to tell users who buys their data. Care and justice: the users are the most vulnerable party and cannot meaningfully refuse. Decision: Sukoon will never sell or share data with advertisers. Aggregated, k-anonymous statistics may go to universities and NGOs, but only with a separate, explicit opt-in. In the presentation, a live audience poll records the room's view before this analysis is revealed.

Part F: Organisational and Operational Structure

F.1 Professional team

StructureStrengthWeaknessFit
FunctionalDeep expertise and clear ownershipSecurity becomes someone else's job; slow hand-offsPartial
MatrixSpecialists keep their standards while delivering for the productDual reporting needs clear rulesChosen
FlatFast for very small teamsUnclear accountability and no independent checkUnsuitable for a high-risk product

In the matrix structure, security, machine-learning, QA and engineering staff report functionally to the Engineering Manager and, for the project, to the Product Owner. The Data Protection Officer and the Ethics Board report directly to the board so that they stay independent of delivery pressure (GDPR Art. 38(3)), and the Ethics Board can veto any release that fails the practice filter funnel. Within our own group of three, one member led engineering, one led security and privacy, and one led ethics and organisation.

F.2 Delivery workflow: Secure SDLC

Every change passes nine gates, modelled in the interactive Secure SDLC pipeline: (1) threat modelling with STRIDE and LINDDUN; (2) peer review; (3) static analysis and secret scanning; (4) dependency scanning and a software bill of materials; (5) automated QA, with 163 unit and integration tests in this repository; (6) dynamic scanning with OWASP ZAP; (7) penetration testing against the OWASP ASVS checklist; (8) a privacy review with a DPIA (GDPR Art. 35); and (9) a release gate with sign-off from the Product Owner and the Ethics Board. A failure at any gate blocks the release.

F.3 Risk register

IDCategoryRiskInherentResidualMitigationOwner
R1LegalNon-compliance with GDPR or the PDPB (consent gaps, unlawful transfer)166Consent ledger, Pakistan-first hosting, SCCs and a TIA, annual auditDPO
R2PrivacyRe-identification and profiling of mental-health data156Data minimisation, redaction, pseudonymous accounts, 30-day retentionPrivacy Engineer
R3CyberBreach followed by dark-web sale of journals208Per-user AES-256-GCM, separate KMS, MFA, web application firewall, penetration tests, a rehearsed runbookSecurity Lead
R4EthicalAI misses crisis messages in Urdu or from women208Balanced data, per-group release gate, human review, clinical advisorsML Lead and Ethics Board
R5OrganisationalRevenue pressure pushes the company towards data selling123Subscription and NGO funding only, funnel gate, Ethics Board vetoEthics Board
R6CyberInsider misuse by a counsellor or administrator124Role-based access, consent-gated access, audited break-glass access, hash-chained logsSecurity Lead

Scores are likelihood multiplied by impact on a five-by-five grid, where 15 or more is critical. Before mitigation, four risks (R1 to R4) are critical; afterwards the highest residual score is 8.

Part G: Research and Industry Validation

G.1 Research

Six peer-reviewed IEEE and ACM papers informed the design; full citations are under References. Research on dark patterns by Gray et al., Mathur et al. and Nouwens et al. shaped our consent interface and our ban on manipulative features. Utz et al. showed how consent notices nudge users, which is why our consent switches carry equal weight and nothing is pre-selected. Narayanan and Shmatikov showed that "anonymised" data can be re-identified, which is why we minimise data and treat claims of anonymisation with caution (risk R2). The survey of bias in machine learning by Mehrabi et al. guided the per-group evaluation in E.1.

G.2 Industry validation

We asked a security professional, Syed Muhammad Saad Bukhari, a VAPT engineer at SNSKIES, to review the privacy and security design. He reviewed the design and architecture; no live penetration test was run. His overall view was that the design is sound for a concept-stage product and that nothing blocks it, but that ten points should be closed before real users join. The most important were to make multi-factor login and short sessions mandatory, to run a full penetration test that includes chatbot abuse, to back up the audit-log hash outside the main system, to require a second approver for emergency access, and to replace rule-only redaction with a trained name-detection step. These are recorded as pre-launch actions, and the full review is in Appendix 2. Review date and sign-off: [[FILL: date and reviewer confirmation]].

Part H: Disclosure and Academic Integrity

Requirements Traceability Matrix

Brief requirementWhere it is met
A: idea, three to five features, architecture, AI and data sourcesPart A; Data-Flow X-Ray
B: privacy audit, protection features, cross-border flow, privacy policyPart B; Appendix 1; Cross-Border Globe; prototype
C: revenue check, PECA trigger, feature audit, mitigationPart C; Attention-Economy Simulator
D: PECA check, dark-web risk, incident responsePart D; Breach Simulator; 72-Hour Incident Clock
E: bias and harm, IEEE and ACM mapping, ethical dilemmaPart E; AI Bias Lens; Ethics Constellation; live poll
F: team, delivery workflow, risk registerPart F; Team Structure; Secure SDLC; Risk Heatmap
G: four to six papers; industry validationPart G; References; Appendix 2
Deliverable 1: reportThis document (Parts A to H)
Deliverable 2: privacy policy and terms of serviceAppendix 1
Deliverable 3: 12 to 15 slide pitch in which every member speaksSlide deck (15 slides, three speakers)
Deliverable 4: individual reflectionsAppendix 3
Bonus: professional practices demonstrated in codeWorking prototype at /app and src/lib, with automated tests
Rules: synthetic data and disclosurePart H

References

  1. C. M. Gray, Y. Kou, B. Battles, J. Hoggatt and A. L. Toombs, "The Dark (Patterns) Side of UX Design," in Proc. ACM CHI, 2018. doi:10.1145/3173574.3174108
  2. A. Mathur et al., "Dark Patterns at Scale: Findings from a Crawl of 11K Shopping Websites," Proc. ACM Hum.-Comput. Interact. (CSCW), vol. 3, art. 81, 2019. doi:10.1145/3359183
  3. M. Nouwens, I. Liccardi, M. Veale, D. Karger and L. Kagal, "Dark Patterns after the GDPR: Scraping Consent Pop-ups and Demonstrating their Influence," in Proc. ACM CHI, 2020. doi:10.1145/3313831.3376321
  4. C. Utz, M. Degeling, S. Fahl, F. Schaub and T. Holz, "(Un)informed Consent: Studying GDPR Consent Notices in the Field," in Proc. ACM CCS, 2019, pp. 973 to 990. doi:10.1145/3319535.3354212
  5. N. Mehrabi, F. Morstatter, N. Saxena, K. Lerman and A. Galstyan, "A Survey on Bias and Fairness in Machine Learning," ACM Computing Surveys, vol. 54, no. 6, art. 115, 2021. doi:10.1145/3457607
  6. A. Narayanan and V. Shmatikov, "Robust De-anonymization of Large Sparse Datasets," in Proc. IEEE Symposium on Security and Privacy, 2008, pp. 111 to 125. doi:10.1109/SP.2008.33

Legal and standards sources: Regulation (EU) 2016/679 (GDPR); Prevention of Electronic Crimes Act 2016 (Pakistan); Personal Data Protection Bill (Pakistan, draft); IEEE Code of Ethics; ACM Code of Ethics and Professional Conduct (2018). The bibliographic details and DOIs of the six papers were checked against ACM, IEEE and university repository records on 17 November 2026.

Appendix 1: Sample Privacy Policy and Terms of Service

This is a sample prepared for a university project. It is not legal advice, and a qualified lawyer must review it before any real use. Last updated: 20 November 2026.

Sukoon Privacy Policy

Who we are. Sukoon is an AI-assisted mental-health first-aid service. You can reach our Data Protection Officer at privacy@sukoon.example (a placeholder address).

What we collect. Only what we need: a pseudonym and your language; your chat messages after we remove names, phone numbers, CNIC numbers, emails, cities and links; the journal entries you choose to save; and your consent choices. We do not collect your contacts, location, advertising identifiers or browsing history.

Why we use it. To reply to you, to recognise a crisis and connect you to a person, to keep your private journal, to let a counsellor help you if you choose, and to keep the service secure. Each purpose has its own switch, every switch starts off, and we only process data for a purpose you have switched on. Because this is sensitive health information, we rely on your explicit consent.

We never sell your data. We do not sell, rent or share your data with advertisers, data brokers or insurers, and we show no advertising. Anonymous, aggregated statistics go to researchers or NGOs only if you opt in separately.

How we protect it. Your journal entries and chats are encrypted with AES-256-GCM using a key that is unique to you and held in a separate key service in Pakistan. Our staff cannot read them. A counsellor can see your journal only if they are assigned to you and you have switched sharing on, or in a documented emergency with a written reason. Every access is recorded in a tamper-evident log, and our code goes through security review, scanning and penetration testing before release.

Your rights. You can access and export your data, correct it, withdraw any consent at any time as easily as you gave it, and erase everything. Erasure deletes your data and destroys your encryption key, so any copies in backups also become unreadable. Use the Privacy Control Center or write to us. You also have the right to complain to the data protection authority.

How long we keep it. Chat messages for 30 days; journal entries for up to 365 days or until you erase them; consent receipts until you erase your account. The security log keeps only pseudonymous identifiers and actions.

Where it is stored. In Pakistan. We send personal data outside Pakistan only with your explicit consent, under standard contractual clauses and with the keys kept in Pakistan. Sensitive data is never sent to countries without adequate safeguards.

If something goes wrong. If a breach affects you, we will tell you in plain language, notify the data protection authority within 72 hours and report the breach to the national cybercrime authority.

Children. Sukoon is for people aged 16 and over, or younger users with a parent's consent where the law allows. We do not knowingly collect data from younger children.

Changes and contact. We will tell you before making any material change and will ask for your consent again where needed. You can contact our Data Protection Officer at the address above.

Sukoon Terms of Service

1. Not an emergency service. Sukoon offers supportive information. It is not a medical service and does not replace professional care. If you or someone else is in immediate danger, contact local emergency services or a crisis helpline now. If our system detects a possible crisis, it may show you emergency resources and, where you have agreed or the law allows, alert a counsellor.

2. Your account. Use a pseudonym, keep your login details safe, and tell us if you suspect unauthorised access. You are responsible for activity under your account.

3. Acceptable use. You agree not to access other people's data or accounts; probe or attack the service without written permission; send spam or unsolicited messages; harass, stalk or threaten anyone, including counsellors; upload unlawful content; or scrape or resell the service. These actions may breach the Prevention of Electronic Crimes Act 2016.

4. Counsellors. Counsellors are independent professionals whose licences we check. They must keep your information confidential and may not contact you outside the platform.

5. Fees. The core service is free. Optional counsellor sessions are paid for by subscription or covered by a sponsor. We will always show the price before you pay, and you may cancel at any time.

6. Our content and your content. You own your journal and your messages. We license the app to you for personal use. Sukoon's software and materials are protected by intellectual property law, and open-source components remain under their own licences.

7. Limits of liability. The service is provided "as is". To the extent the law allows, we are not liable for indirect losses, and nothing in these terms limits liability that cannot lawfully be limited, including liability for death or personal injury caused by negligence.

8. Termination. You may stop using Sukoon and erase your data at any time. We may suspend an account that breaches these terms, after a warning unless the risk is urgent.

9. Governing law. These terms are governed by the laws of Pakistan and disputes go to its courts. This does not remove any rights you have under mandatory consumer or data protection law where you live.

Appendix 2: Industry Validation Record

This appendix records the review referred to in Part G.2. We sent the reviewer the brief below, asked the ten questions that follow, and recorded only what he said.

Brief sent to the reviewer

Subject: A 15-minute review request for a university compliance project.

We are three BSCY students at FAST-NUCES Islamabad taking CS4001, Professional Practices in IT. We have designed Sukoon, a conceptual AI mental-health first-aid app for young people in Pakistan. It is a design exercise that uses synthetic data, and it has no real users. We would value fifteen minutes of your professional opinion on whether its privacy and security design is sound.

The design in brief: anonymous, consent-gated chat in Urdu, Roman Urdu and English; PII redacted before it reaches an AI crisis model; field-level AES-256-GCM encryption with a key per user; hosting in Pakistan first; erasure by deleting rows and destroying the key; role-based counsellor access with audited break-glass access; no advertising and no data selling, with funding from subscriptions and NGOs; a 72-hour incident response plan; and a Secure SDLC that includes penetration testing.

Materials: this report, the sample Privacy Policy and Terms of Service, and the live demo. Your name and organisation will appear in our report only with your agreement.

Questions put to the reviewer

Q1. Is our consent model (granular, opt-in, one purpose at a time, withdrawable and receipted) sufficient for health data under GDPR Art. 9 and the draft PDPB?

Q2. We treat the PDPB as a draft and benchmark against GDPR. Is that a sensible position in Pakistan today, and what should we track?

Q3. Is crypto-shredding, destroying the user's key, an acceptable way to meet the right to erasure when copies exist in backups?

Q4. Are our cross-border rules (Pakistan first, keys in Pakistan, SCCs and a TIA for any transfer) realistic and adequate?

Q5. Which PECA 2016 offences are we most exposed to, and are our controls proportionate? Are the section numbers we cite correct and current?

Q6. Is our incident response timeline (containment within one hour, the authority within 72 hours, users when the risk is high) appropriate, and which authority should we report to in Pakistan?

Q7. From a security point of view, what are the weakest points in our architecture, for example key management, redaction, break-glass access or counsellor access?

Q8. Is our Secure SDLC (threat modelling, static analysis, dependency scanning, dynamic scanning, penetration testing and a DPIA gate) missing anything important?

Q9. Are there legal or ethical risks in an AI crisis detector, such as liability, duty of care or false negatives, that we have underestimated?

Q10. What single change would you make before this design went anywhere near real users?

Response record

FieldEntry
Reviewer name and roleSyed Muhammad Saad Bukhari, VAPT (vulnerability assessment and penetration testing) engineer
OrganisationSNSKIES
Date and format[[FILL: date and format of the review, to be confirmed by the reviewer]]
What was reviewedThe Sukoon design: this report, the architecture and the demo. This was a design and architecture review; no live penetration test was run against any system.
Overall viewThe design is sound for a concept-stage product. Nothing blocks it, but the points below should be closed before real users join.
Changes made as a resultAll ten points below were added to the pre-launch action list. The four strengths listed below were already part of the design.
Points of disagreementNone recorded.

Strengths identified

  1. Encryption by default. Every journal entry and chat message is encrypted with the user's own key, so a stolen database or backup gives an attacker unreadable data.
  2. Consent enforced in code. The app refuses to process anything until the user has switched that purpose on, rather than only describing consent in a policy.
  3. No third parties. There are no advertising networks or trackers, so there is no hidden route for data to leave.
  4. Tamper-evident logs. Every sensitive action is recorded in a hash chain, so any edit to the history can be detected.

Findings and recommendations

#AreaObservationRiskRecommendation
1Key managementEverything depends on the key service. If an administrator can read or export keys, the encryption does not help.MediumUse a managed KMS or HSM, rotate keys, separate administrator duties and alert on every key operation.
2RedactionRule-based redaction catches phone numbers, emails and names after "my name is", but will miss nicknames, school names, misspellings and unusual phrasing.Medium to highAdd a trained name-detection step, hold the message when unsure, and have a person spot-check samples of redacted text.
3Break-glass accessEmergency access is a necessary risk, and a single counsellor could misuse it.MediumRequire a second approver, alert the DPO in real time, limit access by time and review every use afterwards.
4Login and sessionsCredential stuffing and account takeover are the most common attacks on apps like this one.High if missingMandatory MFA for staff and counsellors, passkeys for users, rate limits, lockout and short sessions.
5Audit logA hash chain reveals edits but not the deletion of the newest entries.MediumCopy the latest hash to separate write-once storage at regular intervals and monitor for gaps.
6Browser-side encryptionKeys in the browser are exposed to cross-site scripting, and a lost device can mean lost data.MediumA strict Content Security Policy, no inline scripts, subresource integrity, and a recovery process that does not weaken encryption.
7Erasure and backupsDestroying the key makes backups unreadable only if backup copies of the key are destroyed as well.Low to mediumMake key backups part of the erasure process and test it regularly.
8Testing before launchMost real breaches come from access-control mistakes and chatbot abuse.MediumRun a full penetration test covering API access control, rate limits and input handling, and test the chatbot against prompt injection and attempts to extract other users' data.
9Software supply chainThe project already pins its dependencies and audits them in CI.LowKeep the software bill of materials current and review each new package before adding it.
10Incident readinessA written plan is not the same as a rehearsed one.LowRun a breach drill twice a year and keep enough logs for forensics without storing user content.

Answer to Q10

Complete the full penetration test and fix the login and session findings in point 4 before any real person uses the app.

Sign-off

The reviewer completes this section after reading, editing and approving the record above.

"I reviewed the Sukoon privacy and security design on [[FILL: date]]. In my opinion it [[FILL: reviewer's own wording]]."

Reviewer confirmation: [[FILL: reviewer's name, and signature or email confirmation]]

Appendix 3: Individual Reflections

Each member's statement below covers their contribution to the project and what they learned about professional responsibility, in their own words.

Muhammad Faizan Shakeel (23i-2074)

My contribution: [[FILL]]

What I learned about professional responsibility: [[FILL]]

What I would do differently next time: [[FILL]]

Muhammad Hayyan (23i-2041)

My contribution: [[FILL]]

What I learned about professional responsibility: [[FILL]]

What I would do differently next time: [[FILL]]

Abdul Ahad (23i-2014)

My contribution: [[FILL]]

What I learned about professional responsibility: [[FILL]]

What I would do differently next time: [[FILL]]