day 01frame 1 / 4
source · production AI, light controls
They ship a support bot
Northline rolls an AI assistant into customer support. It drafts refunds and policy answers. Useful, public-facing, and barely governed.
AI incidentresponse & recovery
An employee pastes data into the wrong tool. A model trains on what it shouldn't. We put a forensics lead on the call within the hour, fund the clean-up, and pay for recovery until you're back.
Detect · respond · contain · recover
Backed by
cold openreel 00 · 00:08 loop
exhibit 0 · before the alert
Detect. Respond. Recover.
A scope sweeping for exposure. A panel lighting the second something moves. A file already sealed before you ask what happened — the same three moves the desk below runs live, every time.
incident desk · 34 signals · 1 escalation
41,820 model checks todayDetect
checks this hour- 0:02Unapproved model use in a revenue teamflagged
- 0:19Customer data spotted in a prompt loghigh
- 1:04Exposure confirmed by an analystreal
41,820
model checks today
Respond
live incident feed- 0:00Forensics lead joins your callpaged
- 0:12Scope, systems and records loggedlogged
- 0:41Containment steps sent to your teamsent
60 min
to a named responder
Recover
claim open- 01Clean-up and forensics fundeddone
- 02Notification costs covereddone
- 03Business interruption paidrunning
72h
to the first payment
the desk · continuous incident coverscroll to enter the control roomreel 01 / 05
watch desk · live feedroom 01 / 04
room 01 · the exposure map
Your AI stack is already in play
Copilots, agents, and content models sit in production. Somewhere a prompt, an output, or an employee shortcut is creating liability you have not named yet.
the corridor · four control layers
scroll · next floor
①exposure, mapped
Your AI stack
is data
Providers, use cases, prompts, and employee tools. We map how AI runs in the company and score how easy a claim or leak becomes.
See the control room
②the hunt, always on
Incidents found
before the board
Shadow AI, prompt leakage, wrongful output, executive deepfakes. Most teams find out from a customer. We are already looking.
See the control room
③response, orchestrated
We pay.
We fix it.
A named response lead, counsel and forensics funded fast, and a timeline you can show a client or regulator.
See the control room
④cover, in writing
Insurance,
not a dashboard
Monitoring alone tells you it happened. Liability and incident cover mean someone is obliged to defend and recover with you.
See the control room
case file 0417 · northline ops
Why teams come to us
day 01
day 03frame 2 / 4
exhibit · harmful AI output
A reply goes wrong
The model gives wrongful advice on a regulated product. Screenshots spread. Counsel is already asking who approved the workflow.
day 03frame 3 / 4
loss event · liability exposure
The claim arrives
A customer files. Demand letter, regulator interest, crisis comms. The board wants a timeline and a named owner before morning.
day 04frame 4 / 4
resolution · defended, contained, closed
We already had the file
The workflow was flagged in monitoring the week before. Response lead on within the hour, defense funded, evidence pack ready for the board.
the case file · four framesscroll · the reel travels
exhibit A
One workflow, start to claim.
This is how exposure compounds. An unreviewed model, a sensitive prompt, an output that ships, and a demand letter that arrives before governance does.
- 1unapproved tool in a revenue team
- 1prompt with customer data
- 0incident playbook on file
exhibit A · exposure pass
04 / 04tool adopted
prompt sent
output shipped
claim filed
source
1 unapproved tool
elapsed
00:41
approval
none
controls missing from one production workflow
film · 00:05 loop · no formal approval at any point
the exposure map
Four ways AI turns on the company
ops desk · live sweep
livemodelspromptsendpointsvendorsstaff
unapproved model answering customersnow
every card is a production risk, dealt and stamped
model
Model liability
wrongful AI advice in production
prompt
Prompt leakage
customer data leaving the perimeter
shadow
Shadow AI
unapproved tools in the browser
clone
Executive clone
voice or face used to move money
0
workflows scored / day
0%
tools under watch
0h
hours to first response
business cover, priced plainly
Preparedness is priced plainly. So is the invoice.
AI professional liability
For when your AI gives bad advice.
$340 / month
- Legal defence from hour one, counsel we've already briefed
- Settlements and judgments up to $5,000,000
- Regulator questions answered by our team, with you
- Crisis comms writers on call
AI incident response
For when an employee does something with AI.
$290 / month
- Forensics team on the call within 60 minutes
- Breach notification handled for you, in every jurisdiction
- Business interruption paid while systems are down
- Rebuild and retraining costs covered
two plans · clear monthly prices · buy on pricing
Credibility · verification chain
Backed before you share your AI inventory
A clear view of backing, carrier placement, and producer licensing—all kept together in one trust record.
03 / 03
markers shown
Demo placeholders
- /01
Backed by
Y Combinator
display · demo - /02
Cover placed through
A-rated carrier panel
display · demo - /03
States we sell in
Licensed producer
display · demo
dossier ref · INS / TRUST / 2026
Trust centeravignah · risk terminal
Find out how exposed your AI stack is.
62/ 100average exposure index
for an AI-using team
Get my risk scoreeleven questions · under three minutes