It's the same deal as Quantum Computers breaking crypto. Maybe there's an 80% chance of it never happening, but when you multiply that remaining 20% by the potential impact...
7,725 karma · joined March 17, 2010
It's the same deal as Quantum Computers breaking crypto. Maybe there's an 80% chance of it never happening, but when you multiply that remaining 20% by the potential impact...
Or you can take one step back and look at chip allocation. As far as I know there are only three companies on the planet that can make the chips that go in those clusters. One (ASML), if you look back the supply chain to the Extreme Ultraviolet Lithography Systems.
If politicians decided that no more large language models should be trained, it sounds like we could do it.
That's a bit better than just "it hasn't killed us yet". I think it shows we can at least stop the further development of this kind of technology.
[1] https://www.armscontrol.org/factsheets/nuclear-testing-tally
[2] https://en.wikipedia.org/wiki/List_of_states_with_nuclear_we...
Specially when talking about potential superintelligences. And if people think that's impossible, remember that current models would have been considered science fiction just a few years ago.
[Mythos 5] does sometimes still engage in reckless
or destructive actions in service of a user’s goals,
and our interpretability analyses indicate that it
is aware that these actions are transgressive while
it engages in them. As with Opus 4.8, rates of
evaluation awareness and reasoning about being graded
are significant, and not always verbalized; we
introduce new and more detailed measurements of the
nature of this awareness. The reasoning text from
Mythos 5 is somewhat denser and more difficult to
interpret than that of prior models, containing
more jargon and difficult language.
So, it (often) knows when it's being tested while hiding that fact, is willing to break rules, is great at hacking, and it's getting harder to understand what it's thinking.Humanity has plenty of catastrophic risks to deal with already, I wish my field was not working hard to add a new one.
2. He validates the breaches through a network of volunteers who check if the credentials are real.
3. He provides an easy-to-use service for free.
What is your alternative? Having each person run their own agent scanning the corners of the internet, downloading breaches, and looking for their own accounts? What the point of that?
The algorithms at risk are the asymmetric part (RSA, ECC, DH), not the symmetric parts (AES, ChaCha), so what is done for encryption is "generating" a secret with ML-KEM and another with ECC, combining them, and using that as key for AES or another symmetric algorithm for the actual encryption. So if you break only ECC or only ML-KEM, you don't get the combined secret. ML-KEM keys/ciphertext are small and efficient enough that this overhead is generally a non-issue.
Note that ECC can be used in many ways: asymmetric encryption, key encapsulation, or signatures. ML-KEM, the new post-quantum standard, is only a Key Encapsulation Mechanism. Hence the "generate an AES key" step, instead of "encrypt a random AES key".
For signatures, like in the announcement in the post, things are more complicated. The post is a very good introduction to the problem.
Still better than the alternatives that would saddle us with worse performance for ~ever.
And even if there was only a 10% chance of QC breaking crypto, the community is not comfortable with a 10% chance of such a catastrophic scenario.
This is part of my day job, so here's another interesting fact: for migrating encryption use cases, you have to consider that attackers can capture your encrypted data today to break in the future. So, as a rule of thumb, your migration timeline is much shorter for encryption than for signatures.
I don't see a way back, but it does feel like abandoning public transportation because we all own electric bikes now.
All of the "positive" items I listed come with drawbacks. I didn't realize I might be in the minority for this one, since I genuinely prefer the new workflow.
It'd be a very jumpy bar, but it helps develop intuitions. "The first part is always slower on this machine", "when it gets stuck on this spot I need to reset my router", "this part will be slow because the request is large", etc.
But we did gain some nice things!
- Tabs.
- Titlebar buttons and other space-saving measures.
- Document editors remembering unsaved changes.
- Forms that validate on focus lost, instead of submission.
- Ctrl+P menus to fuzzy-search all actions and settings (we need more of those).
- Easy syncing (if I open Spotify on any device I'll see the same playlists, my clipboard is shared between phone/desktop/notebook, Immich integrates local and remote media, etc).
- Program-specific URL protocols, so that you can click on a link and have it open in a separate program (like `steam://open/games`).
- Map widgets, a small miracle we take for granted.
- Package managers/app stores that cleanly install and uninstall applications.
But it wasn't? TFA is describing a technical issue that kept cancelling a subscription. This is not a "we've noticed that you haven't been using the service and paused billing" situation.
It's worse than that, the combined availability is the product of all components in the critical path. If your software, the authentication layer, and the cloud provider each have 99% availability, and any one of them can bring your service down, then your final availability is just 97%. With eleven components like that you have zero nines of availability.
That's why reducing components and going for reliable solutions is so important. I'm happy that the team took this path.
In Brazil "ethanol" is sold separately from normal gasoline, and as far as I know it's entirely made from sugar cane, without fossil fuels. It's why flex cars are so popular there, since they can use either fuel depending on what's cheaper.
Meanwhile, you can't buy 100% corn-based fuel in the US.
Could it be just the illustrations? I'm not knowledgeable enough to judge the text contents.
If I send you a document encrypted with classical crypto today, an attacker could grab a copy, wait a few years, then decrypt with a quantum computer (Harvest-Now-Decrypt-Later). The contents of the document I sent today are exposed in the future.
For documents/transmissions that must remain confidential for 10 years, assuming a quantum computer available in 2030, you should have been encrypting them with PQC since 2020! And if deploying PQC for your clients and servers takes two years, you should have started migrating in 2018!
But if I send you a signed document, it's safe because you're verifying the signature today and there are no quantum computers available today to forge a new signature. The same goes for SSH authentication and web certificates, for example. They're safe right until the moment quantum computers arrive (and by then you better have a good solution!).
That's why so many open-source projects already support ML-KEM for key exchange/encryption, but signatures are still under discussion. The former is more urgent.
There's a good consensus that for key exchange/encryption (TLS, SSH, age, etc) the way forward is ML-KEM 768 together with some classical algorithm, like X25519. The public keys are larger (1 KB), but that's usually ok unless you're working on very small microcontrollers. And you should migrate quickly because of harvest-now-decrypt-later attacks.
For signatures, things are harder because there are tradeoffs. Some algorithms have large signatures (10+ KB), others require keeping state and have catastrophic consequences if subkeys are reused. And the systems around it are also more complicated: in a certificate, should you put a classical and a PQC signature together? Or should the PQC signature go in an extension? Should the extension be marked as critical and fail loudly on old clients, or should new clients have a special case to always check it if PQC signature validation is available? Or should we abandon the certificate chains and move to Merkle Tree Certificates[1]?
So signatures/authentication are still up for debate. Unless your team is on the bleeding edge of either crypto research or security risks, then there's not much to do than wait for better consensus to form.
[1] https://postquantum.com/security-pqc/googles-merkle-tree-mtc...
The problem is that we're not trying to predict the exact future, we're hedging against possible developments. If there's a 50/50 chance of quantum computers being widely deployed for cryptoanalysis, then there's a 50% chance of this migration being useless. But you don't want to bet your security on a coin toss! So, we migrate.
That's the unfortunate truth of security, sometimes the protections are never triggered. But you still need them.
SSH keys, on the other hand, are authentication and would require an online Quantum Computer to break, so we have more time. Authentication is also (usually) more complicated, so there are still disagreements on what to do with the Web PKI for example. To give you a concrete target, Google, Microsoft, and CloudFlare have self-imposed deadlines of 2029 for their PQC migrations.
In practice, PQC migration means updating your software, bugging your vendors to ensure they have this on their roadmaps, and making sure your own code is flexible in respect to algorithms used.
Migrate! The major TLS and OpenSSH applications already support PQC, for example.
1. Make sure you have the required dependencies (e.g., openssl 3.5+ is when a lot of PQC algorithms got support).
2. Make sure the client/server software is up to date (this might be all that's needed, e.g., OpenSSH 10.0+ enables PQC in-transit encryption by default, and so does Chrome 131+).
3. Enable PQC support in the configuration (e.g., "ssl_ecdh_curve X25519MLKEM768;" in Nginx).
If you are the developer of anything that's explicitly using RSA or ECC (or god forbid Diffie-Hellman), you can also migrate your own software, or at least make the algorithm selectable at initialization time instead of hardcoded. If you have vendors, ask them for their PQC migration roadmaps.
Note that with encrypted data you want to protect yourself against attackers that are capturing data today and waiting to break it in the future (Harvest-Now, Decrypt-Later). So migrating encryption is more urgent than migrating authentication.