up:: The Two-Lane Split
HTTPS vs SMTP Post-Quantum Deployment
What is the difference between web and email post-quantum deployment?
On the same organizations, measured the same day, the web lane runs roughly seven times ahead of the email lane. Published measurement of 4,665 UK organizations found 44.0% of reachable HTTPS endpoints supporting at least one evaluated post-quantum key-exchange group against 6.4% of reachable SMTP services.
The gap is not a difference in organizational effort. Both lanes belong to whoever operates the infrastructure, and the two lanes are usually operated by different providers who made different decisions.
The short version:
- 44.0% HTTPS against 6.4% SMTP, with a matched odds ratio of 16.89 among organizations reachable over both.
- Of 3,510 organizations reachable over both protocols, 144 supported post-quantum on both, 1,419 on the web only, 84 on email only, and 1,863 on neither.
- One provider accounted for 98.4% of all post-quantum-supporting SMTP endpoints in the sample.
- A readiness assessment scoped to HTTPS reports an estate as substantially further along than it is.
- The certificate lane is at zero on both, so neither protocol has begun migrating authentication.
What was measured?
Active scanning on 30 June 2026 across 4,665 UK organizations in 10 sectors, probing TCP port 443 for HTTPS and port 25 for SMTP STARTTLS, with reachable endpoints attributed to their underlying infrastructure providers.
Source: Loizou and Ghadafi, Measuring Post-Quantum TLS Deployment Across UK Internet Sectors, arXiv 2608.02147, 2026.
The two lanes at a glance
| Dimension | HTTPS (web) | SMTP STARTTLS (email) |
|---|---|---|
| Post-quantum key exchange supported | 44.0% (95% Wilson CI 42.5–45.5) | 6.4% (95% Wilson CI 5.6–7.2) |
| Reachable endpoints measured | 4,063 | 3,858 |
| Endpoints supporting | 1,788 | 246 |
| Algorithm diversity | X25519MLKEM768 at 1,787, SecP256r1MLKEM768 at 161, SecP384r1MLKEM1024 at 59, pure MLKEM1024 at 3 | Almost entirely X25519MLKEM768. Exactly one endpoint also supported a second group |
| Provider concentration | Cloudflare: 1,247 of 1,312 endpoints (95.05%). Fastly: 78 of 81 (96.30%) | Google: all 242 of its endpoints. It accounted for 98.4% of all supporting SMTP endpoints |
| Largest non-supporting provider | Several with little or no observable deployment | Microsoft hosted 1,714 reachable SMTP endpoints, none supporting the evaluated groups |
| Provider predicts support | AUC 0.957, against 0.573 for organization sector | AUC 0.994, against 0.156 for sector, which is below chance |
| Post-quantum certificate signatures | Zero observed | Zero observed |
Sources: arXiv 2608.02147, §4. SMTP model figures are the authors’ five-fold cross-validation grouped by anonymized MX host; they also report a row-wise stratified sector-only AUC of 0.699 and explicitly decline to use it as the primary estimate, because it does not account for organizations sharing mail infrastructure.
How do they actually differ?
The web lane was upgraded by content delivery networks and the email lane was not.
A large share of web endpoints sit behind a small number of CDNs, and those providers enabled post-quantum key exchange across their estates. The customer made no decision and got the upgrade. Cloudflare at 95.05% of its measured endpoints and Fastly at 96.30% are that effect.
Email routes through a different, equally concentrated set of providers who made different choices. One provider supported it on every endpoint measured; another hosted 1,714 reachable endpoints with none supporting it. The organizations behind both sets did nothing differently.
⭐ The strongest evidence that this is structural is the sector-only model on the email lane. Once organizations sharing a mail host are accounted for, knowing an organization’s sector predicts its email posture at AUC 0.156, which is worse than guessing. Sector is not a weak predictor there; it is actively misleading, because what it accidentally tracks is which provider that sector tends to use.
Where do they agree?
Both are overwhelmingly one construction. X25519MLKEM768 dominates on both lanes, and pure post-quantum key exchange is close to absent.
Both are provider-determined. The provider-only model beats the organization-only model on both lanes by a wide margin.
Both are at zero on authentication. No post-quantum certificate signature appeared at any endpoint on either protocol, so lane two has not started anywhere in this sample.
And both are measurable from outside using ordinary protocol negotiation and public DNS records, which means an organization’s posture is already visible whether or not it has looked.
Why does the email lane matter more than its share of traffic?
Because of what travels there and how long it stays sensitive.
Email carries correspondence, case coordination, referrals, legal and clinical discussion, and attachments that frequently outlive the systems that produced them. Its confidentiality horizon is long, which is exactly the profile that makes recorded traffic a harvest-now-decrypt-later problem rather than a future one.
And it is the lane almost nobody checks. Assessments and vendor questionnaires gravitate to the web because it is the visible surface.
What does this mean for an assessment?
A posture assessment scoped to HTTPS alone will report an estate as substantially further along than it is. In the measured population, 1,419 organizations supported post-quantum on the web and not on email, against 144 supporting it on both.
The correct scoping question is not “do we support post-quantum” but “on which protocols, operated by whom.”
Establishing the email lane rarely requires a scan. The mail operator for any domain is visible in public DNS records, and the provider’s own posture is a published or measurable property of that provider rather than of the customer.
Common misconceptions
“Our post-quantum score is good, so we are covered.” A good web score is usually the CDN. It says nothing about the email lane, which the same measurement found running roughly seven times behind.
“Email is less sensitive than the web.” Web endpoints mostly serve public content. Email carries the correspondence.
“Our sector is ahead.” On the email lane, sector predicts posture below chance once shared mail infrastructure is accounted for.
“We can fix email ourselves.” Where mail is hosted by a provider, the posture is that provider’s decision. The available action is asking, in writing, and recording the answer.
“Zero certificates means nobody is trying.” The certificate lane depends on ecosystem changes wider than any single operator, and its absence is consistent across every published study.
Questions people ask
Why is email so far behind? Its providers made different deployment decisions than the CDNs did. The organizations are not the variable.
How do I check my own email lane? The mail operator is in public DNS. From there the question is what that operator supports, which is a question to put to the operator.
Is 6.4% getting better? This is a single-day snapshot from 30 June 2026 and cannot establish a trend. Repeating the measurement is what would show movement.
Does this apply outside the UK? The provider concentration effect was corroborated at global scale across a million domains by Wickramasinghe et al., arXiv 2607.29005, 2026. The specific percentages are UK-measured.
What should I ask my email provider? Which post-quantum key-exchange groups their production endpoints support today by name, on what date that changes if none, and whether the answer depends on someone else. See The Vendor-Claims Checklist.
Does hybrid apply the same way on both lanes? Yes, and X25519MLKEM768 dominates both. See Hybrid vs Pure PQC.
If our web is covered, is our harvested traffic safe? Only the traffic that crossed the covered endpoints. Email that traveled under classical key exchange is already recorded if anyone was recording.
Last verified 2026-08-10 · Maintained by Addie LaMarr, LaMarr Labs. Work with Addie at lamarrlabs.com.