Roke folds Vigil AI into platform safety
Public records show the Home Office awarded Roke £1.088m for Vigil AI CAID classifier licences and support from 2026 to 2028. Roke subsequently acquired Vigil AI’s remaining shares, positioning the CAID-trained system for commercial online-safety deployments.
Vigil AI illustrates how sensitive public-sector AI can become commercial trust-and-safety infrastructure, but provenance and governance matter as much as detection performance. Public records describe commercialization of a CAID-trained classifier—not necessarily sale of the underlying archive.
- –The contract covers image and video classifier licences, support, and a redacted attachment marked Official Sensitive.
- –Roke claims Vigil AI detects previously unseen and AI-generated CSAM that hash-matching systems can miss.
- –Chemring says Roke acquired Vigil AI’s remaining shares in May 2026 and secured its first Resolver-channel customer.
- –Platform adoption could support Online Safety Act compliance while reducing moderator exposure to traumatic material.
- –Model-training rights, auditability, false-positive handling, retention, and human review remain critical unanswered questions.
DISCOVERED
2h ago
2026-08-30
PUBLISHED
9h ago
2026-08-29
RELEVANCE
AUTHOR
ThomAsterJourno