Voice assistants and AI: privacy and regulatory issues

Service page: Digital commerce and AI lawyer: terms, privacy and direct mailing.

What happens when you talk to Alexa, Siri or Assistant?

When you say “Hey Siri”, the device wakes up, records what is said, and sends the recording to Apple’s servers. There, the algorithm interprets, responds and sometimes stores it. The same process happens with Alexa, Google Assistant and other similar tools.

This is not a conspiracy theory. It is the technical architecture. And with voice assistants already present in homes, cars, smartwatches and speakers, the amount of information collected about private individuals is enormous. And who keeps all of it? Usually, the company that sold you the device.

What exactly is risky in terms of privacy?

The three main problems:

• Accidental activation (false positives): the wake words (“Alexa”, “OK Google”) are recognized by mistake, and the device records even when you did not mean it to. This has happened, and it is documented.
• Constant listening: to recognize a wake word, the device must always “listen”. Some devices keep voice snippets even without deliberate activation.
• Sharing with third parties: companies passed voice recordings to human reviewers (employees and contractors) for transcription and quality control. Amazon, Google and Apple were exposed for this. The “algorithm improvement” is sometimes done by people who listen to private conversations.

What does Israeli law say?

• Businesses in Israel that use voice AI with customers must, at the least, provide: a notice under section 11 of the Protection of Privacy Law (the purpose of the collection, who will receive the information, and whether providing it is mandatory); informed consent when the collection relies on consent (section 3); and security under the Protection of Privacy Regulations (Data Security), 5777-2017. The right of access is given in Hebrew, Arabic or English (section 13(b)).
• The European GDPR applies to an Israeli business that offers goods or services to people who are in the European Union, or monitors their behavior there (Article 3(2)). Its requirements (a legal basis for all processing, a right of erasure (Article 17) and restriction of processing (Article 18)) are stricter than the Israeli ones.

In practice, consent is included in the terms of use. But an Israeli court can examine whether the consent was “real”: clear and understandable, not just buried on page 12 of the EULA.

The Protection of Privacy Law, 5741-1981, requires anyone who approaches a person to obtain personal information for processing in a database, including voice data, to give that person a notice (section 11). When consent is required, it must be “informed consent” (section 3). This means the user needs to know that they are being recorded, for what purpose, and with whom the information is shared.

What happens when the data leaks?

The Protection of Privacy Regulations (Data Security), 5777-2017, require organizations to secure the database according to the security level that applies to it. In a severe security incident, they must notify the Privacy Protection Authority immediately, and the Authority can order them to notify the affected people as well (regulation 11(d)). A company that was negligent in keeping recordings is exposed to a tort claim (section 31B of the Law) and to a monetary sanction for breaching the Data Security Regulations (section 23Z(h) and the Third Schedule, added by Amendment No. 13).

Voice recordings are especially sensitive: they identify the speaker, reveal habits, and sometimes contain financial or medical information. When the voice is used to identify a person by computer, it is a “biometric identifier”, and therefore “information of special sensitivity” (section 3). A leak of such a database is more serious than a leak of email addresses.

There are significant security risks with voice assistants, including leaks of recordings and hacking

Does a voice assistant listen in order to show ads?

Under Israeli law, using voice data for a purpose other than the one for which it was given (for example, advertising) is an invasion of privacy (section 2(9)). So the purpose must be included in the notice given at the time of collection (section 11) and in the user’s consent. A general, vague consent to terms of use might not be enough.

But even without a microphone, these companies have access to a huge amount of information (location, searches, purchases) that is enough for very accurate predictions. The “magic” of targeted ads does not require listening.

A question that keeps coming up: “I talked about shoes and immediately got ads.” Theory or fact? Officially, there is no confirmation of this. Meta, Google and Amazon have denied using audio from the microphone for advertising, and there is no verified technical evidence.

What can you demand, and how?

If a foreign company refuses to recognize your rights, you can contact the Privacy Protection Authority. In serious cases, a civil claim is also possible.

• Right of access: you can ask to inspect the voice data kept about you in the database (section 13 of the Protection of Privacy Law).
• Right of erasure: you can delete the recording history in Alexa, Google and Siri through the app settings. Under the GDPR this is a legal right, when the grounds in Article 17 exist. In Israel, you can ask to correct or delete information that is not correct, complete, clear or up to date (section 14).
• Limiting the data: you can limit which data is kept, and provide less information.

Users of voice assistants usually have an option to delete recordings in the service settings and to limit the use of their data

Where is regulation heading?

In Israel, the Privacy Protection Authority published, on 28 April 2025, a draft guideline on how the Protection of Privacy Law applies to artificial intelligence systems. There is no specific legislation on voice assistants. Businesses already using voice AI should document what is collected, for how long, and how it is secured. In any case, the database definitions document and the security procedure must record the types of information, the purposes of its use and the security measures (regulations 2 and 4 of the Data Security Regulations).

The EU AI Act classifies remote biometric identification systems (and voice identification can be one) as “high-risk” (Annex III, point 1(a)), except for verification whose only purpose is to confirm that a person is who they claim to be. The use of “real-time” remote biometric identification in public spaces for law enforcement is prohibited, with exceptions (Article 5(1)(h)). High-risk systems are subject to duties of risk management, documentation, transparency and human oversight. According to the European Commission’s page, the application of these duties to Annex III systems was postponed by the Digital Omnibus amendment to 2 December 2027. An Israeli business is exposed to the Act when it markets a system in the EU or when the system’s output is used there (Article 2).