Monthly report
What AI assistants tried to spend
A monthly, signed, k-anonymous report of what apps and AI agents tried to spend through Shield, built only from users who opted in, published only for a month where every cell has at least 25 contributors.
No qualifying month yet. A month is published only when every cell has at least 25 opted-in contributors. Publishing a page of suppressed cells would be theatre, so there is none. The first month will appear here, signed, when it qualifies.
Method
- Only people who opted in. Telemetry is off by default. A contribution is a coarse category and threat vector — a renewal held, a data leak blocked — keyed by a salted one-way hash. Never an amount, never a merchant, never who.
- k = 25. Every published cell has at least 25 distinct contributors. Cells below that publish nothing — not even the count. No grand totals, no cross-tabs, so a suppressed cell cannot be recovered by subtraction.
- Tests are excluded. The in-app test charge and the agent demo never contribute.
- Signed and frozen. Each month is one immutable JSON artifact, signed with a dedicated Ed25519 key whose public half is published separately; it can be verified offline. The signature proves integrity, not truth; k-anonymity limits small-group disclosure, it does not guarantee anonymity against outside knowledge.
- No money estimates, no prevented-loss claims. Counts of what was attempted and what was held, by category.