SSH scan finds post-quantum key exchange support on 45% of sampled hosts

A snapshot of SSH readiness

Researchers Md Ariful Haque, Sander Vinberg, Ben Edwards and João Cruz of Bitsight Technologies examined which key exchange algorithms public-facing SSH servers advertised. SSH provides secure remote access; key exchange lets two parties establish a shared secret over an insecure network.

Their study reports post-quantum support on 45.2% of sampled hosts. At the organizational level, 39.9% of the 27,012 mapped entities had at least one supporting host. These are different measures: an organization qualifies for the second statistic even if most of its observed servers lack support.

The paper was submitted to arXiv on November 6, 2026. It is a preprint; the supplied record does not establish peer-reviewed publication. Its measurements describe a February 16, 2026 snapshot, not deployment conditions at submission time. This article draws on the full paper and its versioned arXiv record.

How the researchers measured support

The team used GROMA, Bitsight’s proprietary internet-scanning platform, to collect SSH records on port 22 for one scan day. After cleaning approximately 1.55 million records, they retained 1,523,998 unique IP addresses with available key exchange information.

They searched the advertised algorithm lists using a custom regular expression designed to cover post-quantum algorithms listed by the Open Quantum Safe project. A host counted as supporting post-quantum cryptography if its list contained at least one matching algorithm. Classical and post-quantum options could coexist on the same server.

Bitsight’s AI-powered GIA attribution engine mapped IP addresses to organizations. The researchers then combined those mappings with industry, regional, revenue, employee and digital-footprint information to compare deployment patterns.

This method measures advertised capability. It does not establish which algorithm actual clients negotiate, how frequently they use post-quantum options, or whether authentication also uses post-quantum signatures. The reported percentages therefore should not be read as the share of fully quantum-secure SSH connections.

Hybrid algorithms dominate

The scan identified 688,814 hosts advertising at least one matching post-quantum algorithm, versus 835,184 without one.

The most common listed variant was sntrup761x25519-sha512@openssh.com, with 681,490 hosts. It combines Streamlined NTRU Prime with the classical X25519 mechanism. The related name without the vendor suffix appeared on 168,414 hosts. The paper explains that these names refer to essentially the same algorithm under different naming conventions.

Table II contains a numerical inconsistency: its percentages do not consistently match its counts and the stated sample size. Dividing those two counts by 1,523,998 gives approximately 44.72% and 11.05%, rather than the table’s 44.86% and 11.08%. The leading ML-KEM variant, mlkem768x25519-sha256, appeared on 88,088 hosts, approximately 5.78% using that same denominator, rather than the reported 5.79%.

Other observed variants were much rarer. Algorithm counts overlap because one server can advertise multiple options; adding them would double-count hosts.

The authors suggest that OpenSSH software upgrades and default settings may explain some of the widespread support. That is a possible explanation, not something the scan establishes about administrators’ intentions.

Organizations show uneven coverage

Of the mapped organizations, 10,788 had support on at least one observed SSH host, while 16,224 had none. Looking within organizations revealed a polarized distribution: roughly 60% had support on less than 1% of their observed hosts, while around 23% had near-total coverage.

Near-total coverage here means coverage of the measured SSH key exchange capability, not completion of an organization-wide cryptographic migration.

Education and telecommunications led the sector comparisons, with more than half of entities showing support somewhere in their observed SSH infrastructure. Technology had the largest absolute count of supporting entities but ranked third by percentage.

The researchers also found similar revenue and employee-count distributions among supporting and nonsupporting organizations. Larger digital footprints were associated with adoption. Their scatter plot shows fewer entities with complete measured coverage at higher IPv4 counts, but does not establish that the proportion achieving complete coverage declines with size. These descriptive comparisons do not establish that budget is irrelevant or that network complexity causes incomplete deployment.

What the snapshot cannot establish

Geographic comparisons covered only the 15,820 entities with regional information, rather than the full organizational sample. The paper reports relatively higher adoption in North America and Western Europe, but missing regional data limits the scope of those comparisons.

The underlying data are proprietary, limiting independent reproduction from the supplied material. The study also covers observed public-facing SSH services on port 22, not every SSH deployment or internal server.

Finally, a single day’s observations cannot measure migration speed or demonstrate a trend. The strongest conclusion is narrower: post-quantum key exchange capability was already widespread in this sample, while its presence and coverage varied substantially across organizations.