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 LH028

LH028 rapid checkpoint: 22 October 2026, 16:00 UTC slot

Actual direct read: 16:00:05 to 16:00:14 UTC. Panel: the three URLs fixed in the brief, within amendment 1. Raw record: snapshot, replayed with python3 studies/LH028/scripts/replay.py --snapshot 2026-10-22T1600Z. Compared with the 13:42 UTC initial snapshot. One direct GET per source; no additional source pages read.

Source Access at this read Raw-byte comparison Visible article-text comparison
Amazon Threat Intelligence HTTP 200, 195-byte Site Unavailable placeholder, SHA-256 5b131ca14aa96311d3432b0062c7443d3b0e6346ec279bd376a75f0a76bcd5d7 Identical to initial placeholder No article text available to this collector
Cloudflare HTTP 200, 301,730 bytes, SHA-256 3d0f90b29a5d79e364b79ce217eb2d05906662e0d6128efb076690839ac16f86 Changed from d007f9a68aa3b868ff732f65450324bb40c4ed1127519a046893b5b201a79b37; first difference is a script nonce Identical in ordered HTML text chunks after excluding script, style and SVG content and normalising whitespace
OpenAI HTTP 200, 454,210 bytes, SHA-256 9f514ba974c936a9bd32c8184c6bc828b4738c12dd6074226f6b15a387c19ae0 Changed from 8e1a95952ed6170d0019f1a78c62971731fbc684ee214b9d6bb3d71374cb4d95; first difference is script/hydration material Identical under the same visible-text comparison

The changed page hashes do not establish revised claims. This comparison found no changed visible article text in the two accessible bodies, and the AWS direct access gap persists. HTML text extraction is a check on the retained pages, not a rendered-browser inspection; hidden attributes or scripts may differ. No new event-specific numerator, denominator, unit or observation window is available from this read. The earlier source-reported attempted exploitation, Cloudflare's defensive deployment, and OpenAI's documented defensive AI role therefore remain separate; this checkpoint adds no evidence of successful compromise or attacker AI use. No public article is revised on this checkpoint.

From studies/LH028/checkpoints/2026-10-22T1600Z.md in the repository, last changed 22 October 2026.