Dev.to Security ๐Ÿ” Cybersecurity ๐Ÿ‘ 0 ๐Ÿ“– 7 min read

Post-Quantum Authentication Is the Hard Half of the Migration

Key exchange got all the attention. Last week, Azure started rolling out hybrid post-quantum key exchange for App Service, and AWS shipped it in DocumentDB 8.0.3. Browsers negotiate ML-KEM in the background now, and nobo

Key exchange got all the attention. Last week, Azure started rolling out hybrid post-quantum key exchange for App Service, and AWS shipped it in DocumentDB 8.0.3. Browsers negotiate ML-KEM in the background now, and nobody has to think about it. It looks like the migration is basically done.

It is not. The people actually deploying this stuff met in Prague last month, and the Internet Architecture Board published their first impressions on 12 November. The message from that room is blunt: key exchange was the easy half. Authentication is where the migration gets slow, expensive, and genuinely hard.

I have been building with post-quantum signatures for a while now. My messenger uses ML-DSA-65 identity keys, the same algorithm at the center of this story, so I read the workshop notes the way a builder reads them: what breaks, what it costs, and what I would have to change in my own code. Here is what Prague made clear.

The size problem is not a footnote

An Ed25519 signature is 64 bytes. An ML-DSA-65 signature is 3,309 bytes. That is not a rounding difference; it is a different size class, and it breaks limits that were set decades ago and never revisited.

The UK Government Digital Service measured what happens to a normal identity verification journey when you swap in ML-DSA-65 signatures and ML-KEM-768 key exchange. The data that journey sends grew from 6.7 KB to 44.5 KB. A signed OAuth request that travels in a URL now has to fit inside limits that differ at every hop: 8 KiB at Amazon CloudFront, 16 KiB for a load balancer's request line. Nobody designed those limits with 3 KB signatures in mind, because nobody had to.

Cisco ran the same question through SD-WAN simulations. At 3% packet loss, 16.3% of ML-DSA-87 handshakes over DTLS 1.2 took more than five seconds. Over DTLS 1.3, that number was 0.2%. The difference is that DTLS 1.2 retransmits a whole flight of handshake messages when one datagram disappears, and a post-quantum handshake is a lot of datagrams. Sean Turner, who has spent years inside these standards, put it plainly in the session: "Please don't do DTLS 1.2 and PQ."

That is the shape of this migration. The cryptography works. The transport underneath it was not built for it, and the transport has to change first.

Verifiers move first, and they move slowest

Here is the part that changed how I think about the whole migration. In key exchange, both sides negotiate and move together. In authentication, the verifier has to move first, and verifiers are the slowest thing in the system.

A signer cannot switch algorithms until every verifier that checks its signatures can verify the new ones. And the migration is not finished when you deploy new certificates. It is finished when verifiers stop accepting the old ones. Google's David Benjamin stated the goal in his position paper: an attacker with a quantum computer must not be able to intercept connections between updated clients and updated services. If that is not achieved, he wrote, "PQ certs just waste bytes." An attacker picks which certificate to present, so a client that would still accept a forged classical certificate is still vulnerable, no matter what the real server deployed.

The timelines in Prague were sobering. npm said it would probably wait at least 12 months for people to adopt a CLI that can verify post-quantum signatures before it starts offering them. Let's Encrypt still issues RSA certificates from a separate hierarchy because some clients cannot validate ECDSA, years after ECDSA became normal. Aaron Gable from Let's Encrypt said issuing from the ECDSA hierarchy would "break more than 1% of our customers," and called that "the unfortunate reality of how poorly the transition to ECDSA has gone."

And then there are the devices that cannot be updated at all. Cisco's SD-WAN presenters said the classical certificates built into their routers at the factory may stay in use until about 2035, because replacing all that hardware by 2030 is not possible. As long as those certificates stay in use, an attacker who can forge a classical certificate could impersonate a network controller and take over the network. Some devices must verify firmware against the same root for 20 to 30 years after installation. Netflix's paper gave this a name: forge-later attacks, where attackers use quantum computing to forge authentication in systems that rely on long-term trust.

Root rollover makes it worse. When the old classical root signs a new post-quantum root, trust still rests on the algorithm being retired. As Mike Ounsworth put it in Prague: "The thing that you're rooting is the thing that's about to die." There is no flag day when everyone switches at once, because clients update their trust anchors at different times.

Every shortcut moves the cost somewhere else

The workshop looked at the proposed workarounds, and each one trades one constraint for another.

Composite signatures combine ML-DSA with a classical signature, so both have to validate. On Day 1 they were described as the fix for "I don't trust the cryptography." Ounsworth said composites "turn a zero-day into a 20-day." But Benjamin pointed out that a public PKI is likelier to fail by trusting the wrong party, and signing twice does not fix that unless two certificate authorities attest to the same thing.

Merkle Tree Certificates shrink what goes on the wire by letting one signature cover many objects, with an inclusion proof tying each one to what was signed. The design needs many hashes but few signatures, and Gable said internet scale is not the hard part. The hard part is the details: the certificate log must have a single writer and never fork, few hardware security modules can sign with ML-DSA today, and no enrollment protocol can deliver a second certificate for the same request at a later time. No library yet builds the new certificate's signature block.

Then there is the simplest lever of all, and it came from Netflix: stop signing things that do not need signing. Where nobody needs to prove to a third party who sent a message, Netflix prefers keys derived from a shared secret, using HMAC or authenticated encryption, over signatures. Its paper says the move to post-quantum authentication "is not a substitution problem." The price, the Netflix presenter admitted, was complexity: the backend used to treat every type of device the same way, and it no longer can.

The IAB co-chairs made the same point in their own words: before choosing a signature algorithm, decide what actually needs signing. In my own messenger, that is a question I keep coming back to. Every signature in the protocol has a reason, or it should. The ones that exist only because signing felt like the safe default are the ones I should look at first.

The standards are splitting

The fourth obstacle is the one nobody in the room could solve: different jurisdictions are pulling in different directions. Some deployments will need pure post-quantum algorithms, others will require composites, and size-constrained applications may need entirely different constructions like Merkle trees. One participant put it plainly: "The NSA says no hybrid while somebody else requires it." A vendor serving many countries pays for each conflicting policy.

And then there was the surprise guest. AI coding agents came up as an unexpected cause of divergence. One participant described a deployment where coding agents had no single authoritative source for which post-quantum algorithm and implementation to use, so different agents implemented different formats. Agents also write their own cryptography instead of calling a library, which works against crypto agility, the ability to change algorithms later. The suggestions in the room included scanning pull requests for hand-written cryptography and writing guides that coding agents can read. That is a strange new failure mode for a migration: not a broken algorithm, but a dozen agents each confidently implementing a different one.

What this means for builders

The workshop's full report is still coming, and the co-chairs were careful to say these are first impressions, not guidance. But the direction is clear enough to act on.

First, measure what you actually send. The IAB's advice is practical: size your deployment against every limit on the path, handshakes and requests alike, before you pick a signature algorithm. The transport layer may need to change before the credential formats do.

Second, list the verifiers you cannot update. They determine how long the old algorithm stays trusted, and your plan has to say when you stop accepting it. That is the definition of done now, and it is the part most migration plans skip.

Third, question every signature. If no third party needs to check it later, a shared secret and a MAC may be the better answer, and it is certainly the smaller one.

And finally, do not confuse deploying post-quantum certificates with being post-quantum secure. The key exchange half of the migration is nearly invisible now, and that is a real achievement. But the authentication half is where trust actually lives: in roots, in verifiers, in devices that will still be running in 2035. That half will take longer, cost more, and demand more judgment than the first. Prague was the first time the people doing the work said so out loud, with measurements to back it up. It is worth listening.

๐Ÿ“ฐ Read the original article on Dev.to Security

Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes โ€” full credit and traffic to the original publisher.