Product surface
Every AI change in one place. Already documented.
What a regulated team opens on Monday: everything their AI changed last week, sealed without anyone writing a word — plus what is about to go wrong, and the fix for it.
The estate below is synthetic — a live one would be a customer's private data, and the names are *.example per RFC 2606. Everything computed from it is not: the risk model, its per-feature explanations, the obligation coverage and the audit export are the production code paths running in your browser. The model weights are priors set from the failure modes the architecture is designed around — they are not fitted on customer data, because there is none yet.
Acme Financial Group
8 governed workflows · 6 providers
Overview
Every AI change across the estate, scored before it bites, with the evidence already sealed.
91%
58 of 64 changes carry sealed, offline-verifiable evidence.
5 unattested · 3 approvals overdue
Changes, last 7 days
37
High or critical risk
18
28% of the feed
Approvals outstanding
19
3 past SLA
Time returned
12days
of manual change-record work removed
Change activity
Daily volume by change type, last 21 days
- Prompt
- Model
- Params
- Tools
- Policy
› Table view
| Day | Prompt | Model | Params | Tools | Policy | Total |
|---|---|---|---|---|---|---|
| 11 Jul | 1 | 1 | 0 | 0 | 0 | 2 |
| 12 Jul | 2 | 1 | 0 | 0 | 0 | 3 |
| 13 Jul | 0 | 0 | 0 | 0 | 0 | 0 |
| 14 Jul | 0 | 1 | 0 | 0 | 0 | 1 |
| 15 Jul | 2 | 0 | 0 | 0 | 0 | 2 |
| 16 Jul | 1 | 0 | 0 | 1 | 0 | 2 |
| 17 Jul | 1 | 0 | 0 | 0 | 0 | 1 |
| 18 Jul | 1 | 0 | 0 | 0 | 1 | 2 |
| 19 Jul | 0 | 0 | 1 | 1 | 1 | 3 |
| 20 Jul | 0 | 1 | 0 | 0 | 0 | 1 |
| 21 Jul | 0 | 1 | 0 | 0 | 1 | 2 |
| 22 Jul | 0 | 1 | 0 | 0 | 1 | 2 |
| 23 Jul | 0 | 1 | 0 | 0 | 0 | 1 |
| 24 Jul | 3 | 0 | 0 | 2 | 0 | 5 |
| 25 Jul | 2 | 0 | 0 | 0 | 0 | 2 |
| 26 Jul | 1 | 0 | 1 | 1 | 0 | 3 |
| 27 Jul | 2 | 1 | 1 | 0 | 0 | 4 |
| 28 Jul | 5 | 1 | 1 | 0 | 0 | 7 |
| 29 Jul | 0 | 0 | 0 | 0 | 1 | 1 |
| 30 Jul | 5 | 3 | 0 | 2 | 0 | 10 |
| 31 Jul | 3 | 2 | 3 | 2 | 0 | 10 |
Risk distribution
Modelled probability of an incident or audit finding
- Critical5
- High13
- Elevated17
- Low29
› Table view
| Segment | Value | Share |
|---|---|---|
| Critical | 5 | 7.8% |
| High | 13 | 20.3% |
| Elevated | 17 | 26.6% |
| Low | 29 | 45.3% |
What is driving estate risk
Every change's feature contributions, summed across the feed
- Sensitive data in scope30.3%
- Wide blast radius18.2%
- Agent authority widened11.4%
- Approvals missing10.3%
- Model swapped8.7%
- Provenance incomplete8.3%
› Table view
| Item | Share of total risk push | Detail |
|---|---|---|
| Sensitive data in scope | 30.3 | % |
| Wide blast radius | 18.2 | % |
| Agent authority widened | 11.4 | % |
| Approvals missing | 10.3 | % |
| Model swapped | 8.7 | % |
| Provenance incomplete | 8.3 | % |
Provider mix
Which vendors are serving governed traffic
- Self-hosted1422%
- In-house1320%
- Anthropic1117%
- Google1016%
- Mistral813%
- OpenAI813%
› Table view
| Item | Changes | Detail |
|---|---|---|
| Self-hosted | 14 | 22% |
| In-house | 13 | 20% |
| Anthropic | 11 | 17% |
| 10 | 16% | |
| Mistral | 8 | 13% |
| OpenAI | 8 | 13% |