Post-quantum TLS adoption remains uneven beyond the web

Measuring deployment rather than performance

A preprint submitted to arXiv on November 6, 2026, by Bitsight researchers Md Ariful Haque, Sander Vinberg, Ben Edwards and João Cruz, examines post-quantum cryptography deployment across seven TLS-based services. Its findings should be read as preprint research rather than established conclusions.

The motivation is “harvest now, decrypt later”: an attacker could retain encrypted communications until a sufficiently powerful quantum computer can break the classical public-key mechanisms protecting them. Post-quantum key exchange is intended to address that future threat. This study measures negotiated key exchange, however, not every component required for fully quantum-resistant communications.

Unlike work benchmarking algorithm speed, the researchers probe deployed services. Their results suggest that adoption is concentrated in HTTPS and secure directory services, with much lower rates elsewhere.

How the scans worked

Using Bitsight’s GROMA internet-scanning platform, the team assembled 10,000 IPv4 targets for each service: HTTPS, LDAPS, SMTPS, IMAP-SSL, POP3-SSL, TELNETS and RDP. These cover web access, directories, email and remote administration.

They scanned from a Google Cloud instance using OpenSSL 3.5 and an internally developed tool. Custom extraction rules read handshake traces to identify the negotiated TLS key-exchange group. Of 70,000 service-target scans, 5,517 were unsuccessful, mainly because connections were refused or timed out. Subsequent analyses excluded those failures.

The team also attributed addresses to organizations using Bitsight’s GIA asset-mapping system. Revenue and employee information came from service providers, while internal datasets supplied digital-footprint measures. This enabled organizational comparisons, but makes those results dependent on attribution and business-data quality.

HTTPS and LDAPS lead the sample

Among successful connections, the reported proportions negotiating post-quantum key exchange were:

• HTTPS: 9.11%.
• LDAPS: 7.39%.
• IMAP-SSL: 0.64%.
• POP3-SSL: 0.51%.
• SMTPS: 0.45%.
• TELNETS: 0.15%.
• RDP: 0.01%.

For HTTPS, that meant 816 positive results among 8,960 successful connections. RDP produced just one positive result among 8,602.

The only post-quantum group observed was X25519MLKEM768, a hybrid combining classical X25519 with the post-quantum ML-KEM mechanism. It accounted for 2.6% of successful connections across all seven services. Classical X25519 remained the most common negotiated group, at approximately 49.3%. Another 43.2% were classified by the authors as not supporting TLS 1.3.

These are observations under the scanner’s handshake configuration. They do not establish every algorithm a server could support, or what proportion of ordinary users’ sessions actually negotiate post-quantum protection.

Organizational adoption is not complete migration

Across 4,636 mapped organizations, 308 had at least one sampled endpoint supporting the observed hybrid group. Within each service’s organizational sample, adoption reached 11.04% for HTTPS and 5.42% for LDAPS. The remaining services ranged from 4% for IMAP-SSL down to 0.11% for RDP.

The threshold matters: one positive endpoint qualifies an organization as an adopter. It does not mean its entire infrastructure has migrated. Likewise, 100% coverage among an organization’s sampled endpoints cannot establish coverage of assets outside the scan.

For HTTPS, technology had the highest reported sector adoption rate, around 17%. Media and entertainment, telecommunications, and transportation exceeded 10%, while several sectors showed near-zero adoption in the sample.

HTTPS adopters had somewhat higher median revenue and employee counts, and a more noticeably elevated median IPv4 count. Larger address footprints also showed slightly lower rates of complete sampled deployment. These descriptive associations do not prove the authors’ proposed explanations involving budgets, staffing or migration complexity.

Why the Cloudflare comparison needs caution

A separate scan tested 11,979 Cloudflare-hosted domains using Server Name Indication, which tells a TLS server which hostname the client wants. The paper reports 9,297 negotiations using X25519MLKEM768, alongside a claimed adoption rate of 83.6%. Those figures are inconsistent: 9,297 divided by 11,979 is approximately 77.6%.

A separate Cloudflare-associated IP scan found 437 positive results among 11,048 targets, roughly 4%. Neither endpoint measure is directly interchangeable with the Cloudflare Radar figure cited by the authors: 57.3% of human HTTPS traffic. Domain counts, IP counts and traffic-weighted measurements have different denominators and connection conditions.

A useful snapshot, not a readiness census

The paper does not establish a representative random sample of the internet or clearly specify the scanning dates. It covers selected public-facing IPv4 services, excludes failed connections and cannot describe unobserved internal systems. Its conclusion also incorrectly groups LDAPS below 5%, despite reporting 7.39% in the results.

The strongest supported takeaway is narrower: in these scans, hybrid post-quantum key exchange was substantially more common on HTTPS and LDAPS than on the five other services. Visible web progress should not be mistaken for organization-wide migration.