I found an academic article (Vogel et al 2019) that analyses this phrase. Key points:

• when the German legislator uses geschäftsmäßig, this demonstrates a clear difference in intention from gewerbsmäßig or gewohnheitsmäßig

• the article quotes Franz von Liszt 1881, and this definition seems to be accepted to this day:

  Die Gewerbsmäßigkeit charakteriſiert ſich einerſeits durch die auf öftere Wiederholung gerichtete Abſicht, andrerſeits durch die Abſicht des Thäters, ſich durch dieſe Wiederholung eine, wenn auch nicht regelmäßig oder dauernd fließende Einnahmsquelle zu verſchaffen […].

  Die Geſchäftsmäßigkeit teilt mit der Gewerbsmäßigkeit die auf regelmäßige Wiederholung gerichtete Abſicht, dagegen fehlt die Abſicht, ſich eine ſtändige Einnahmsquelle zu eröffnen. Ob die einzelnen Handlungen honoriert werden oder nicht, iſt gleichgültig.

• the term geschäftsmäßig is significant for §5 TMG, but has also reached wider attention in the discussion around the decriminalization of assisted suicide.

So the key defining aspect is the auf regelmäßige Wiederholung gerichtete Absicht, the intention directed towards regular repetition.

This meaning in legalese German is divorced from everyday language.

§ 5 TMG has the interesting construction of “geschäftsmäßige, in der Regel gegen Entgelt angebotene Telemedien”. So the TMG does not seem to care whether you have a profit motive, only that other people might provide this kind of service for a profit motive. If other people would provide instances of Discord bots in order to get donations, that might already bring you in scope.

This is not legal advice, but it seems like your options are to either avoid running an instance of the bot, only running it in a private context without access from a wider public, or sucking it up and providing the necessary documentation.

And no, it is probably not possible to use a PO box because you don’t live or work at that address. The general expectation seems to be for the address in an imprint to be ladungsfähig, so that you can be served there. This random lawyer’s website writes:

Unter der Anschrift in diesem Zusammenhang ist die Postleitzahl, der Ort, die Straße und die Hausnummer zu verstehen, nicht ausreichend ist die Angabe eines Postfachs.

That’s not the correct criterion. There are multiple German laws that require imprint-style disclosures.

Some of them are indeed specific to commercial activities.

But the Impressumspflicht typically means §5 TMG which requires an Impressum for

geschäftsmäßige, in der Regel gegen Entgelt angebotene Telemedien

Rough English translation:

Telemedia offered in a business-like manner, typically for remuneration

Critically, “geschäftsmäßig” does not mean “commercial” or “profit-oriented”. In particular, nonprofit organizations also act geschäftsmäßig.

IANAL, but it doesn’t sound like your service wouldn’t be geschäftsmäßig.

All of this is irrelevant anyway because you very likely have to publish a privacy notice per Art 13 or Art 14 GDPR. This must include the identity and contact details of the data controller (i.e., you). The German data protection authorities expect that the identity includes your real name and a ladungsfähige Anschrift (address where you can be served), so pretty much exactly what would be included in an Impressum anyway.

For a project like Signal, there are competing aspects of security:

• privacy and anonymity: keep as little identifiable information around as possible. This can be a life or death thing under repressive governments.

• safety and anti-abuse: reliably block bad actors such as spammers, and make it possible for users to reliably block specific people (e.g. a creepy stalker). This is really important for Signal to have a chance at mass appeal (which in turn makes it less suspicious to have Signal installed).

Phone number verification is the state of the art approach to make it more expensive for bad actors to create thousands of burner accounts, at the cost of preventing fully anonymous participation (depending on the difficulty of getting a prepaid SIM in your country).

Signal points out that sending verification SMS is actually one of its largest cost centers, currently accounting for 6M USD out of their 14M USD infrastructure budget: https://signal.org/blog/signal-is-expensive/

I’m sure they would be thrilled if there were cheaper anti-abuse measures.

This article is ahistoric and unnecessarily conspirational.

Signal and its predecessors like TextSecure have been run by different companies/organizations:

• Whisper Systems
• Open Whisper Systems
• Signal Technology Foundation (and its subsidiary Signal Messenger LLC)

Open Whisper Systems received about 3M USD total from the US government via the Open Technology Fund for the purpose of technology development … during 2013 to 2016. Source: archive of the OTF website: https://web.archive.org/web/20221015073552/https://www.opentech.fund/results/supported-projects/open-whisper-systems/

The Signal Foundation (founded 2018) was started by an 105M USD interest free loan from Brian Acton, known for co-founding WhatsApp and selling it to Facebook (now Meta).

So important key insights:

• It doesn’t seem like the Signal Foundation received US government funding. (Though I haven’t checked financial statements.)
• The US government funding seems to be a thing of the fairly distant past (2016). The article makes it sound like the funding was just pulled this year.
• The US government funding was small compared to Signal’s current annual budget. It was not small at the time, but now Signal regularly makes more from licensing its technology than it regularly received from the US government. According to ProPublica, Signals financial statements for 2022 indicate revenue of about 26M USD

It would be unwise for a bank to publish its exact fraud detection and risk management policies, otherwise they could be easily circumvented. A lot of these policies will be embodied in their internal backend services.

Someone will now inevitably mention “security by obscurity”. But Kerckhoff’s Principle is specifically about cryptosystems which should derive their security solely from the strength of the keys. That way, confidentiality is still ensured even when details about the cryptosystem become known to adversaries.

But non-cryptographic aspects of security benefit from asymmetric knowledge, from grey areas, from increasing risk for adversaries.

Cryptography works. At least until sufficiently powerful quantum computers arrive, TLS reliably ensures confidentiality between your browser and the server. No one else can snoop on the data transmitted via that connection.

But are you connected to the right server? Without some kind of authentication, any adversary in the middle (such as your ISP) could impersonate the real server.

That is where certificates come in. They are issued by neutral certificate authorities (CAs) that check the identity. It works something like this:

• I, the server operator, create a private key on that server. I use that key to create a certificate request which asks the CA to give me a certificate. This request also contains the domain names for which the key shall be used.
• The CA performs identity checks.
• The CA issues me the certificate. I install it on my server. Now, when browsers create a TLS connection I can tell them: here’s my public key you can use to check my identity, and here’s a certificate that shows that this is a valid key for this domain name!
• The browser will validate the certificate and see if the domain name matches one of the names in the certificate.

What kind of checks are done depends on the CA. I’ve obtained certificates by appearing in person at a counter, showing my government ID, and filling out a form. Nowadays more common is the ACME protocol which enables automated certificate issuance. With ACME, the CA connects to the server from multiple network locations (making interception unlikely) and checks if the server provides a certain authentication token.

To know which certificates are valid, browsers must know which CAs are trusted. Browser makers and CAs have come together to create an evolving standard of minimum requirements that CAs must fulfill to be eligible for inclusion in the browser’s default trust store. If a CA violates this (for example by creating certificates that can be used for government traffic interception, or by creating a certificate without announcing it in a public transparency list), then future browser versions will remove them, making all their certificates worthless.

eIDAS 2 has the effect of circumventing all of this. There is to be a government-controlled CA (already high-risk) that has its own verification rules set by legislation (does not meet industry standard rules). And browsers would be legally forced to include the eIDAS CAs as “trusted”.

This puts browsers in a tough spot because they’ve resisted these kinds of requests from authoritarian regimes in the past. But now the world’s largest trade bloc is asking. Browsers can comply or leave the EU market, or maybe provide a less secure EU edition? Awakens uncomfortable memories around the failed US attempts at cryptography export control (cryptography is considered a munition, like hand grenades or ballistic missiles).

It is plausible that the EU is doing this with good intentions: having a digital identity scheme is useful, it makes sense for identity to be government-controlled (like passports), and such a scheme will only see success if it sees industry adoption. The EU has also seen that hoping for voluntary industry adoption doesn’t generally work, e.g. see the USB-C mandate.

Well, it’s about Peter Thiel, who also founded the Palantir surveillance technology company. As a source for his involvement with Brave, Wikipedia cites this TechCrunch article, which mentions funding from Thiel’s “Founders Fund”.

I’d rather criticize Brave for other reasons though, like being led by Brendan Eich or supporting crypto.

C++ does have the problem that references are not objects, which introduces many subtle issues. For example, you cannot use a type like std::vector, so that templated code will often have to invoke std::remove_reference and so on. Rust opts for a more consistent data model, but then introduces auto-deref (and the Deref trait) to get about the same usability C++ has with references and operator->. Note that C++ will implicitly chain operator-> calls until a plain pointer is reached, whereas Rust will stop dereferencing once a type with a matching method/field is found. Having deep knowledge of both languages, I’m not convinced that C++ features “straightforward consistency” here…