Privacy and Surveillance
AI runs on data, and much of that data is about people: what we post, what we type into chatbots, our faces and our voices. This page explains what is collected, where the law draws lines, the rights you have — in Europe and in Malaysia — and what you can do about it.
In January 2020 a New York Times reporter, Kashmir Hill, revealed that a little-known company called Clearview AI had quietly built a searchable database of billions of faces scraped from the public internet — social media, news sites, personal blogs — and was selling access to police. Upload one photo of a stranger and, within seconds, it could return other pictures of them and the pages they came from. Nobody in those photos had been asked. That story became a test case on three continents, and it captures the privacy question in the age of AI: if something is technically "public", may anyone collect it, learn from it, and use it to find you?
This page explains data protection law for general understanding. It is not legal advice. Check the current law and official guidance — in Malaysia, the Personal Data Protection Department (JPDP) — before relying on it.
What AI systems collect
Personal data enters AI systems at several layers. Open each one, from the outside in.
Your chats are data too
People confide in chatbots — about health, relationships, work disputes — as if talking to a diary. A diary does not sit on someone else's server. Three facts are worth knowing.
Training defaults vary. Most major assistants let you switch off the use of your conversations for training, but whether it is on or off by default differs between providers, between free and paid plans, and between personal and business accounts, and policies change As of 2026. Business and enterprise plans usually exclude training by contract. Check the settings of each tool you use; the Novice page Staying Safe with AI shows where to look and what never to share.
Deleting is not always deleting. Providers keep data for safety review and legal reasons. In 2025, as part of the New York Times copyright case against OpenAI, a US court ordered OpenAI to preserve chat logs that users had deleted, as potential evidence; the order was later narrowed. It showed that a legal case you are not part of can affect what happens to your chats (see Copyright and Creativity).
Regulators are watching. Italy's data protection authority, the Garante, temporarily blocked ChatGPT in March 2023 and in December 2024 fined OpenAI €15 million over the legal basis for training on users' data, transparency and age checks. OpenAI appealed, and according to the Garante's own website a Rome court upheld the appeal in March 2026 As of 2026. In May 2025 Meta began training on public posts by adult users in the EU after offering an objection form — a move privacy campaigners such as noyb criticised as opt-out rather than consent.
Faces in the crowd
Facial recognition turns a face into a numerical template and compares it with others. Used one-to-one ("is this the phone's owner?") it is convenient and fairly benign. Used one-to-many — searching a face against a huge database, or scanning crowds live — it becomes a tool of identification at a distance, and the stakes change. It has also been shown to make more errors on some groups: US government (NIST) testing in 2019 found higher false-match rates for many algorithms on Asian and African-American faces, though the best systems have improved since.
Clearview AI became the test case. Here is how regulators in different places responded As of October 2026.
Data protection authorities in France, Italy and Greece each fined Clearview €20 million in 2022 and ordered it to delete their residents' data. In 2024 the Dutch authority fined it €30.5 million, said its database of more than 30 billion images was illegal under the GDPR, and warned that Dutch organisations using the service could themselves be fined. Clearview, which has no EU office, has disputed that European law applies to it and has reportedly not paid most of these fines. The EU AI Act has, since February 2025, banned building facial-recognition databases by untargeted scraping of images from the internet or CCTV.
The Information Commissioner's Office fined Clearview £7.5 million in 2022. In 2023 a tribunal overturned the fine on the ground that Clearview's clients were foreign police, outside UK law's reach. In October 2025 the Upper Tribunal reversed that, holding the regulator did have jurisdiction because Clearview was monitoring people in the UK, and sent the case back for the merits to be decided. The fine itself had not been reinstated at the time of writing.
In October 2021 Australia's Information Commissioner found that Clearview breached the Privacy Act by collecting Australians' facial images without consent, and ordered it to stop and delete them. The Administrative Appeals Tribunal largely upheld that finding in 2023. In August 2024 the Commissioner announced it would not pursue further action, while noting the determination still stands — a reminder that winning a ruling and enforcing it against an overseas company are different things.
Surveillance or safety?
Debated Police forces in many countries use facial recognition to find suspects and missing people; shops use it against repeat shoplifters; schools and employers use AI to monitor attention or productivity. Supporters see a powerful safety tool. Critics see the end of being anonymous in public.
Should police be allowed to use live facial recognition in public places?
Choose one to hear its case.
The case
Cameras are already everywhere; matching faces against a short watch-list of wanted people is faster and cheaper than human officers studying footage. With limits — a published watch-list policy, deletion of non-matches within seconds, human checking before any arrest, independent audits — it can catch serious offenders and find vulnerable missing people. The EU AI Act takes roughly this line: real-time remote identification by police in public is banned except for narrowly defined serious cases with prior authorisation.
The weak point
Safeguards depend on institutions behaving well; watch-lists tend to grow; errors fall unevenly; and the chilling effect on protest and ordinary life is hard to measure but real.
The case
Being anonymous in a crowd is a basic freedom. Once every face can be identified, people may avoid protests, places of worship or clinics. Mistaken matches have led to wrongful arrests — most of the documented cases in the United States involved Black people. Several US cities, starting with San Francisco in 2019, banned police use outright.
The weak point
A blanket ban gives up real benefits, may simply push use into private companies or less accountable agencies, and ignores that the technology's accuracy keeps improving.
- Civil-liberties groups (such as Amnesty International, Big Brother Watch in the UK and the ACLU in the US): ban mass facial recognition in public spaces altogether.
- Many police forces and some governments: it is a proportionate tool when tightly governed; the UK's Metropolitan Police, for example, has expanded live deployments, citing arrests of wanted suspects.
- Regulators taking a middle line: the EU allows narrow police exceptions with judicial authorisation; data protection laws like the GDPR and Malaysia's amended PDPA treat biometric data as sensitive, requiring stronger justification.
- Technologists point out that accuracy and bias depend heavily on the specific system, camera quality and threshold settings, so blanket claims in either direction can mislead.
Your rights: the GDPR and Malaysia's PDPA
Data protection laws give individuals rights against organisations that hold their data. The European Union's GDPR (applying since 2018) is the global reference point. Malaysia's Personal Data Protection Act 2010 covers personal data in commercial transactions — not the Federal and State Governments — and was significantly amended in 2024, with changes in force in stages during 2025. The basics of the PDPA and its seven principles for organisations are on AI at Work; here the focus is on what you can ask for As of 2026.
SEEAccessFind out what they hold
GDPR: you can ask any organisation for a copy of your personal data, the purposes, who it was shared with and how long it will be kept; usually free, answered within one month. PDPA: you can make a data access request in writing; the organisation must normally respond within 21 days and may charge a small prescribed fee.
FIXCorrectionMake it accurate
Both laws let you ask for inaccurate, incomplete or out-of-date data to be corrected. This matters for AI: a wrong record can feed a credit-scoring or fraud model. Note that chatbots can also invent false "facts" about real people; European complaints (for example by noyb in 2024) argue the GDPR's accuracy duty applies to such outputs.
DELErasure and objectionMake them stop
GDPR: a "right to be forgotten" in many circumstances, and a right to object to processing based on the company's "legitimate interests" — the basis many AI firms rely on for training. PDPA: no general right of erasure, but you can withdraw consent, require an organisation to stop processing that is likely to cause you substantial damage or distress, and stop direct marketing.
MOVEPortabilityTake it elsewhere
The GDPR lets you receive data you provided in a machine-readable format and move it to another provider. The 2024 PDPA amendment added a similar right of data portability, in force from 1 June 2025, subject to technical feasibility and format compatibility; detailed guidance was still developing As of 2026.
!Breach notificationBeing told when it goes wrong
GDPR: organisations must tell the regulator within 72 hours of becoming aware of a breach, and tell you directly when the risk to you is high. PDPA (from 1 June 2025): data controllers must notify the Commissioner as soon as practicable — the Commissioner's guideline sets 72 hours — and notify affected individuals without unnecessary delay where significant harm is likely.
BOTAutomated decisionsA human in the loop
The GDPR gives a right not to be subject to decisions based solely on automated processing that significantly affect you (such as an automatic loan refusal), with exceptions, plus a right to human review. The PDPA has no equivalent provision; Malaysia's approach to AI decisions has so far been through voluntary guidelines As of 2026 — see Regulating AI.
Opting out of training sits across these rights. In the EU, where AI firms rely on "legitimate interests", you have a right to object, which is why Meta and others offer objection forms there. In Malaysia and most other countries there is no general legal right to stop public posts being used for training, so in practice your levers are the provider's own settings, keeping accounts private, and — for anything you share with a company directly — withdrawing consent.
Children's privacy
Children generate data long before they can understand consent: school apps, games, parents' photos, and now chatbots marketed as tutors or companions. Most laws treat them as needing extra protection. The GDPR lets countries set an age, between 13 and 16, below which parental consent is needed for online services; the US Children's Online Privacy Protection Act (COPPA) covers under-13s; the UK's Age Appropriate Design Code (2021) requires services likely to be used by children to default to high privacy. In September 2025 the US Federal Trade Commission opened an inquiry into how AI "companion" chatbots affect children and teenagers.
Malaysia's government announced in 2025 that it intended to bar under-16s from holding their own social media accounts, with platforms to verify age using official identity documents (eKYC) under the Online Safety Act framework; implementation was still being worked out in 2026 As of 2026. Age verification itself raises privacy questions: proving you are over 16 can mean handing more identity data to platforms. For practical family advice, see the children's section of Staying Safe with AI.
Practical steps
You cannot control everything, but a few habits remove most of the avoidable risk. Start at the top.
- Share less in the first placeDo not put IC numbers, bank details, medical records or other people's private information into chatbots. Anonymise examples before asking for help.
- Check the training and history settingsIn each AI tool, find the data controls: turn off training if you prefer, use temporary or incognito chats for sensitive topics, and delete old conversations you no longer need.
- Lock down what is publicMake social accounts private where you can, remove old photos you would not want scraped, and think twice before posting children's faces with names, school uniforms or locations.
- Protect your face and voiceBe wary of free apps that ask for selfies or voice samples. Agree a family "safe word" against voice-clone scams.
- Use your legal rightsSend a data access request to a company you suspect holds your data; ask for corrections; withdraw consent for marketing. In the EU, use objection forms for AI training.
- Complain when it goes wrongIn Malaysia, complaints about how a company handles personal data go to the Personal Data Protection Department (JPDP); scams go to the National Scam Response Centre on 997.
A Malaysian online shop leaks its customers' data and the leak is likely to cause significant harm. Under the amended PDPA, who must it tell?
Since 1 June 2025, breaches must be notified to the Commissioner, and to individuals where significant harm is likely.
Why did the UK's 2023 tribunal first overturn the fine on Clearview AI?
The Upper Tribunal reversed that view in 2025, holding the regulator did have jurisdiction.
// Pause and reflect
- Which AI tools do you use most? Do you know whether your conversations are used for training — and would you change that setting now?
- Would you accept live facial recognition at a shopping mall or train station in exchange for better safety? What conditions would you insist on?
- Think of one company that holds a lot of your data. What would you want to see if you sent it a data access request?
Your notes are saved privately in this browser.
Key terms on this page
Sources & further reading
- European Union, Regulation (EU) 2016/679 (General Data Protection Regulation), Articles 15–22 and 33–34 (2016).
- Personal Data Protection Department (JPDP), official website: Personal Data Protection Act 2010, the 2024 amendments and guidelines (accessed October 2026).
- Baker McKenzie, “Malaysia: Personal Data Protection (Amendment) Act 2024 to come into force” (2025).
- Office of the Australian Information Commissioner, “Statement on Clearview AI” (August 2024).
- Hunton Andrews Kurth, “Dutch regulator fines Clearview AI 30.5 million euros” (2024).
- A&O Shearman, “Upper Tribunal finds in favour of ICO in appeal … on Clearview AI” (October 2025).
- Lewis Silkin, “OpenAI faces €15 million fine as the Italian Garante strikes again” (2025).
- Patrick Grother et al., NIST, Face Recognition Vendor Test, Part 3: Demographic Effects (NISTIR 8280) (2019).