Pavel Durov discusses Telegram’s engineering culture, ownership, privacy choices and business model with Lex Fridman. He also recounts conflicts with authorities and explains his personal philosophy. Seven research notes examine the evidence behind security claims, French procedural terminology, product design, subscriptions, the Gram settlement and a mouse experiment used as a human analogy. Private financial figures and disputed personal or political accounts remain attributed to the speakers.
HostLex Fridman
GuestPavel Durov
About these timings. Publisher-supplied paragraph starts explicitly linked to the YouTube recording; paragraph ends are unknown and words may occur later. Exact 4:34:10 duration verified from the matching official YouTube result. Podcast audio contains a different introduction and its timings are not used.
Durov describes Telegram’s encrypted cloud and says it protects messages from employee and government access.
Background and context
The March 2023 full version of Albrecht and colleagues’ security study distinguishes default cloud chats, encrypted between client and server, from optional end-to-end encrypted one-to-one Secret Chats. These designs place trust at different boundaries. A claim about encrypted storage is therefore not equivalent to a claim that only the communicating users possess access to message content.
The researchers investigated older implementations, reported vulnerabilities to Telegram in April 2021, and recorded Telegram’s confirmation that specified client versions addressed them. Some attacks were practical; others depended on difficult assumptions. Their positive proof concerns a modified protocol model under stated assumptions. This is neither evidence that the 2025 service remains vulnerable to those attacks nor an independent verification of Durov’s sweeping claim that private messages have never leaked.
An investigation is a procedural status, not a verdict
Why this point. Procedural terminology prevents both an accusation being mistaken for guilt and the guest’s criticism being mistaken for a court finding.
Durov says that he is being investigated by a French investigating judge and is not on trial.
Background and context
The Paris prosecutor’s statement of 28 August 2024 records a mise en examen and judicial supervision. The publicly indexed official excerpts include a five-million-euro security deposit, reporting requirements and an initial restriction on leaving France. These are historical procedural facts, not proof that the allegations were established at trial. The complete PDF was unavailable during this research, so the note relies only on those identifiable excerpts.
France’s official procedural explanation, verified in January 2026, describes a mise en examen as an investigating judge’s decision within a judicial inquiry that carries defence rights and obligations. That later general explanation helps translate the term; it does not establish the current outcome of this particular case. Neither source verifies Durov’s separate allegations of political pressure or resolves whether the investigation was justified.
Durov describes animated stickers as a combination of vector artwork, compact files and smooth motion.
Background and context
Telegram’s July 2019 launch announcement specifies a Lottie-based TGS format for animated stickers. It reports typical files around 20–30 kilobytes and animation at 60 frames per second. Vector instructions describe shapes and movement instead of storing every frame as a full image, which explains the design choice behind the interview’s comparison with heavier animation formats.
The announcement is useful primary evidence of what Telegram launched and how it described the format. It is still the vendor’s account. Its claims about bandwidth, battery use and performance are not an independent benchmark across all devices, and the format alone cannot establish that every sticker or competing implementation performs equally well. The product achievement combines a constrained format with artwork, rendering code and device testing.
Durov says reproducible builds let people check that Telegram’s distributed apps match its published source code.
Background and context
Telegram announced reproducible Android and iOS builds with version 5.13 on 31 December 2019. The relevant check is whether an independently built client corresponds to the app distributed through the stores. This makes the relationship between an inspectable codebase and a shipped executable testable, rather than merely trusting that they match.
A matching build does not, by itself, establish that the source contains no security flaws, that a remote server behaves as claimed, or that the product is safer than every competitor. Those questions require different evidence. No build reproduction was performed for this companion. The dated announcement supports the existence and purpose of the verification process, while the interview’s broader comparative security claims remain attributed to the speakers.
Durov describes Premium as paid extra features that preserve the free service, and gives later subscription and revenue figures.
Background and context
The June 2022 Premium announcement sets out the original bargain: existing features remain free while subscribers obtain additions such as larger uploads, higher limits and extra tools. Some benefits extend to non-subscribers, including receiving larger uploaded files. The announcement also says Premium payments help support the free service. This provides a dated primary record of the intended business model.
It is not an audited financial statement. The launch document cannot establish the subscriber total, revenue forecast or profitability discussed in September 2025. Those figures remain Durov’s or Fridman’s statements in the conversation. Nor does an announcement of paid features establish the company’s costs, retention or how much each revenue stream contributed. Evaluating those business outcomes would require financial and operational records for the relevant period.
The Gram settlement concerned a particular securities offering
Why this point. The contemporary settlement record gives concrete legal and financial terms without turning them into a general judgment on cryptocurrency.
Durov says Telegram abandoned its original blockchain launch after a conflict with the SEC over the fundraising.
Background and context
The SEC’s announcement of 26 June 2020 describes court-approved settlements with Telegram Group and TON Issuer over the unregistered Gram offering. It reports an order for $1.224 billion in disgorgement, with credit for repayments to initial purchasers, and an $18.5 million civil penalty. The defendants settled without admitting or denying the complaint’s allegations.
This is a regulator’s primary account of a particular offering and settlement. It does not say that all blockchain software was prohibited, and it does not independently establish every step in the later community project’s development. It also cannot validate claims about the present network’s speed, security or investment value. Keeping the original Gram fundraising dispute separate from later uses of TON makes the interview’s compressed history easier to assess.
Fridman and Durov use Calhoun’s mouse experiment to discuss abundance, motivation and possible social decline.
Background and context
Calhoun’s January 1973 paper describes four breeding pairs placed in a closed enclosure with plentiful food, water and nesting space. The mice could not leave. Population growth, social organisation and reproduction changed together; the enclosure was not an experiment that varied wealth alone while holding every other social condition constant.
The paper reports a peak near 2,200 mice and, at its final editing, 27 survivors. It discusses the expected extinction of the population, so an eventual endpoint should not be confused with an observation already completed when the article was written. Calhoun himself proposed human analogies, but the subjects and controlled environment remain mice in a bounded habitat. The study does not establish that human prosperity, a particular inheritance policy or future AI abundance necessarily causes social collapse.
Remember a word from the conversation? Find an approximate listening time without reading through a transcript.
Transcript paragraph starts; matching words may occur later. All times are approximate. Use up to six words; matching uses 30-second windows, not exact phrases. Transcription can be wrong, so listen to check. Searches stay in your browser.
PodCompanions