Lighthouse

Lighthouse watches the internet for what AI systems do there and what their activity leaves behind, and publishes what it finds, with the evidence and its limits.

Studies LH042

How do published counts of internet-exposed AI services move over time, read from the scanners' and researchers' own publications?

In short. We gathered the counts that security researchers and scanning companies have published of AI services left reachable from the internet, kept only those that say how, when and what they counted, and dated each by its scan. There are 91 such counts from 19 publishers, and they disagree by large factors, partly because they count different things: on one day in May 2026 one writer got about 1,200, 25,500 and 30,000 exposed Ollama servers from three ways of asking two scanning indexes. Contrary to what our horizon sweep found, counts after February 2026 do exist for both Ollama and OpenClaw; since February the only run made one way is one research group's monthly index built from Shodan, whose OpenClaw category (which it says need not be OpenClaw itself) fell from about 267,000 in April to about 154,000 in late June and stood near 206,000 on 1 November. These are the publishers' claims, checked against their own pages and not against any host; reachable is not compromised, and Lighthouse scanned nothing. We'll read that index again in December and January.

Study, version 0.2, 8 November 2026 studies/LH042/LH042.md

From studies/LH042/ in the repository, last changed 9 November 2026.