https://protonmail.com/blog/protonmail-threat-model/
Which says don't use it if you are up against state actors and: "Sensitive business communications – You have sensitive business information that you want to make sure is protected from competitors and other malicious parties. For example, you fear a competitor may want to sue you under false pretenses to get access to sensitive data. In this case, ProtonMail offers a great deal of protection. ProtonMail will not release ANY data unless provided with an enforceable Swiss court order. To get such an order, the case must first work its way through the Swiss courts where stricter privacy laws might result in a different ruling. Even if an adversary went through the expensive and time consuming procedure of obtaining such an order, ProtonMail’s zero access cryptography means we would only be able to release data that is encrypted since we do NOT hold the decryption keys."
Given that they have Javascript to handle the decryption keys, couldn't they demand ProtonMail change the code delivered to your browser session to give up the keys? This would make the only extra security provided by this scheme would be the multiple court jurisdictions and the less tested legality of a court order making their product less secure, ala the FBI and Apple.
> "Nearly every country in the world has laws governing lawful interception of electronic communications. In Switzerland, these regulations are set out in the Swiss Federal Act on the Surveillance of Postal and Telecommunications Traffic (SPTT) last revised in 2012. In the SPTT, the obligation to provide the technical means for lawful interception is imposed only on Internet access providers, so ProtonMail, as a mere Internet application provider, is completely exempt from the SPTT’s scope of application. This means that under Swiss law, ProtonMail cannot be compelled to backdoor our secure email system."
[0]https://protonmail.com/support/knowledge-base/protonmails-ss...
Why bother?
If they can MITM ProtonMail, they might as well use letsencrypt which just requires you control the domain name (for some definition of control).
> I'm not sure what the state of certificate pinning is,
Public-Key-Pins-Report-Only: pin-sha256="Jh0ZSUYEc97HRYcwZIOkH2jKOUpmQhLO48MYd1s5pRM="; pin-sha256="2ZnCTNQBrKShr4c1olKfwNG53KiL6qoNcQi65YGRBn8="; pin-sha256="1D76xWwHug9p4iQWVBiDZF+e3UcxtPte/ig5pkYnmRU="; max-age=300; report-uri="https://protonmail.com/pkp-report"
This is often used as an argument by EV advocates, but it doesn't hold up under scrutiny. An attacker with access to a non-EV certificate can selectively intercept only connections for subresources of the targeted site (i.e. JavaScript). The "main" connection would still use the EV certificate and thus show the browser indicator. This attack was first made public in 2008[1] and has been further refined in later work[2].
HPKP and the Expect-CT header provide some viable mitigations for this. That said, it seems unlikely to me that a nation-state adversary would choose to attack at the Web PKI level in this scenario. Compromising ProtonMail or the user's device would probably cheaper and less likely to be detected.
[1]: http://w2spconf.com/2008/papers/s2p1.pdf
[2]: https://www.blackhat.com/presentations/bh-usa-09/SOTIROV/BHU...
Switzerland is not an island of privacy with regard to state surveillance – and with regard to private data privacy, it basically mirrors the European Union’s standard. According to Snowden documents, Swiss intelligence and security services are close partners with the NSA and other foreign services.
Do you think that's partly because, like most countries, the truth is... obfuscated.
In other words, the law does not say that ProtonMail is exempt for having to provide lawful interception. They might still have to do so, based on some other law.
Further, it's quite conceivable that unlawful means (such as blackmail, threats, or bribery) could be used to coerce ProtonMail. That's not to mention perfectly lawful means of enticing them -- like appealing to their patriotism, willingness to help in a critical investigation, or demonstrating some credible threat.
Swiss authorities and security services cooperate closely with partners all over the world including the NSA. And there is a longstanding and working network of mutual legal assistance including the Convention on Cybercrime (CCC).
A lot of people in our industry and others have this same model. It is not that the NSA is after you. It's local law enforcement. In our case, SEC and FBI. For average users, local LE will not have 0 days or anything special. Protonmail is out of reach for a lot of LE and that is important.
The best choice for us when we're fully operational is to run a Protonmail-like setup, self-hosted. Deal with mail issues and spam. It is a pain!
Arent there many (difficult to judge how many) cases of people losing access to their Google account, and therefore about everything they had online (photos, email, videos, etc...). That is also scary enough, especially when it happens randomly with no clear reason why and the support of Google seems to be limited to sending info via forms in the hope of a future human interaction.
If a service has thousands of times more users than another, you'll expect weird edge cases to show up thousands of times more often.
People underestimate the power of a calm, deliberate letter sent by post with an elected representative or two copied.
It shows you're serious. Most people complain to blow off steam. This is what customer service handles. If you want something done at the corporate policy level, e.g. to have policy changed or have a decision made per policy reversed, you attack at higher levels.
The traditional form of leverage is legal. Have a lawyer pen a letter gets you out of customer service. It is as effective as it is expensive. Next best is a regulator (you can think of these as narrowly-scoped, publicly-funded lawyers). Unfortunately, nobody regulates Google.
So your final threat is getting a lawmaker pissed off with you. This is less about passing legislation (it's hard to pass legislation; everyone knows that) than creating an official, reliably-corroborated paper trail which could go public, causing PR damage, and/or damage relationships the company may want to lean on in the future.
More practically, I don't want to sit around writing and responding to letters. Having someone else do the back and forth with me Cc'd is more pleasant.
Disclaimer: I am not a lawyer. This is not legal advice. If you need legal advice, contact a lawyer.
There are, but that's not incorporated into a threat model because it's not a security issue (at least not a first-order security problem).
Unauthorised access, content modification or deletion, impersonation, and several other categories of security policy violations are also fairly typical.
Most of the cases of people losing access to their Google account that I've seen are not ones which would could feasibly be induced by a dedicated attacker unless they already had access to your account in the first place.
That may be a part of your personal threat model and also your decision-making, but from Google's perspective, that's not a security issue.
Yes, you could argue that only the loud ones making noise get their accounts back, which my be true, but considering that billions of people use GMail, the fact that so few stories even show up means that it's probably safer than even air travel.
To my knowledge, it was Flame that did this not Stuxnet, and it was an attack on MD-5, not SHA-1 [1]
I thought employees do not have access to user data. Can anyone comment on this?
With months of effort researching tripwires and auditing systems, any of them could read your mail.
There's a pretty good chance they'd get caught by some auditing or alerting system they were unaware of though. Many of those systems are kept secret from employees for obvious reasons.
Any two employees collude to much more easily read your mail. There's probably ~1000 people in that position (not only the gmail team, but anyone who writes any kind of library code used by any of the databases, datastores, or application servers). They would leave audit records though, although they might go unnoticed.
A convenient way to put a target on your back. What benefit does this have to their security? Accountability?
I expect that any single Google employee does not have access, in the sense that they've locked themselves out of making changes to their systems without a computer seeing multiple human approvals. But there is a point at which someone can fix that computer.
But yes, my point was that the threat model here isn't "one Google employee goes rogue," it's "multiple Google employees with extremely privileged access to Gmail and/or their code deployment servers all decide that they want to read your email without telling you, and nobody blows any whistles," which seems much more far-fetched.
Really? You've heard just as many stories of unexpected Google account closures as I have.
I'm quite confident ProtonMail just don't do that. And if they did, you'd have a much more credible chance of talking to a human and rectifying the situation.
But I think there are lots of other options besides ProtonMail that will do the same. (I pay $50/year for Pobox.com, personally, and vaguely feel like paying a bit of money for my email instead of relying on a free service is worthwhile.)
It's abused fairly regularly in fact.
There are a lot of cases but I'm at work and can't spend too much time collecting much. The gist however is that they are bound by Swiss speech laws, which are very ambiguous. "Inciting violence" is one such example, but of course, they can't see what you are supposedly inciting, so they suspend you.
There are some people for whom "The government is literally after me, personally" is a valid threat model. There are some people for whom "Google employees with privileged access to Gmail are conspiring to be after me, personally" (one assumes there's a two-person rule for access to individual inboxes or deploying code that scans inboxes) is also a valid threat model.
However, those people should consider that the government will be willing to use either software 0-days or algorithmic 0-days to attack them (see e.g. Stuxnet taking out Iran's nuclear program using a previously unknown method of generating SHA-1 collisions, that looked kind of like how the academic community knew to generate collisions but with a different fingerprint), in the government case. Or that any interaction with anyone who uses Google must be avoided, in the Google case. See e.g. https://mako.cc/copyrighteous/google-has-most-of-my-email-be...
For normal people (which includes me and probably everyone else commenting here)? Google seems at least as likely, probably a tiny bit more, to protect me from threats like "A personal relationship has gone bad and someone who isn't a government and isn't Google is trying to impersonate be me" or "I don't want to lose access to my email" (remember that availability is a part of security!).