The public record of what actually goes wrong.
SkyeWatch ingests two public registers — CISA’s Known Exploited Vulnerabilities catalogue and the AI Incident Database — then filters and classifies them for sovereignty relevance. Not vendor marketing about hypothetical risk. The incidents that were actually exploited, and the AI failures that were actually reported.
What the feeds are carrying right now.
Pulled from the running platform. If a source goes quiet, this page says so rather than presenting stale rows as current.
Classified, not just collected.
Every incident is filtered for sovereignty relevance and given a cluster, a severity and a one-line summary. Classification runs on the sovereign model — incident text never leaves to a third party.
Loading the register…
Why we run it
Sovereignty arguments are usually made in the abstract. SkyeWatch exists to keep ours anchored to the public record: what was actually exploited, where data actually went, and which AI systems actually failed — rather than a threat model assembled to flatter the product.
The sovereignty filter is deliberately wide. An incident is kept if it touches privacy and security or matches a keyword set covering compelled disclosure, extraterritorial reach, residency, egress, lock-in, key access, training data and model exfiltration. False positives are cheap; a missed incident is not.
Where the data comes from
- CISA Known Exploited Vulnerabilities — the US authority’s catalogue of vulnerabilities with confirmed exploitation in the wild. Not theoretical CVEs; the ones being used.
- AI Incident Database — the public register of reported AI harms and failures, maintained independently of any vendor.
Both are public registers, published by third parties, and cited with links back to the source. SkyeWatch adds the sovereignty lens, not the facts.