Stealth, but not secretive.
blkbx is a stealth startup building evidence infrastructure for insurable AI. Not because the idea is hidden — because the category is still forming.
The company began with a simple observation: AI systems are becoming operational actors before the insurance market has the evidence infrastructure required to underwrite them.
Every major industry wants to use AI agents. Every insurer wants better visibility into AI risk. Every broker wants a clearer way to package AI exposure. Every enterprise wants to prove that its AI controls worked.
But the current evidence layer is broken — scattered across logs, dashboards, model cards, governance reports, screenshots, emails, and incident reviews. blkbx replaces fragmented post-hoc reconstruction with machine-generated evidence receipts at the moment of action.
Our view is simple
- 01
If an AI action can create a loss, it should create a receipt.
- 02
If it creates a receipt, it can be reviewed.
- 03
If it can be reviewed, it can be priced.
- 04
If it can be priced, it can become insurable.
Our principles.
Record actions, not vibes.
AI trust should not depend on a slide deck. Every consequential action should produce evidence.
Evidence must be portable.
A receipt should be reviewable by enterprises, brokers, insurers, reinsurers, auditors, and regulators.
Logs are not enough.
Logs preserve chronology. Receipts preserve meaning, authority, controls, and reconstructability.
Reconstructability is measurable.
The question is not whether a system claims to be governed. It is whether its actions can be reconstructed after a loss.
Obstruction is signal.
When evidence fragments fail to agree, that inconsistency should be surfaced, not hidden.
Insurance needs a machine-readable evidence layer.
The future of AI insurance will not be priced only from questionnaires. It will be priced from receipts.
We are currently working with founders, insurers, brokers, technical collaborators, and enterprise AI teams who believe the same thing: AI risk will not become bankable, insurable, or governable until AI actions generate evidence by default.
The category is early · The thesis is public · The recorder is being built