Put a price on every AI risk you run

The Risk Agent estimates the annual loss from each threat on each AI system in euros, and repeats the estimate on your schedule. When people build new agents, the figure keeps up with them.

Request a demo Explore a sample portfolio

A sample portfolio of 1,200 AI systems across 12 business units. Every Monday at 09:00 the Risk Agent prices each threat on each system. Half of the exposure sits in 162 systems, and the largest source is Forecast helper agent 2, an agent in Claims found by discovery, at €3.0M a year.

Your AI portfolio grew faster than your risk register

Once thousands of people can build agents with a low-code licence, new AI systems appear every week. Most risk programmes still answer with a heat map that a workshop fills in once or twice a year.

The heat map sorts risks into high, medium and low, but it cannot tell the board what the exposure is in money, and it is out of date the week after the workshop.

Find the agents that carry most of your exposure

Each AI system gets an estimate for every threat that applies to it, from prompt injection to a vendor changing its model. Line up every AI system by exposure and a short head carries most of the money, while the long tail can be reviewed in bulk. Some of the largest sources are agents that nobody registered, which discovery found.

Explore a sample portfolio

1Forecast helper agent 2Claims · found by discovery€3.0M
2Reserving notebook agentActuarial · registered AI system€1.5M
3Onboarding helper agent 4Underwriting · found by discovery€1.4M

Every figure shows its working

The Risk Agent builds each estimate the way an actuary sketches one on paper: how often the incident happens, what one incident costs, and how much of it your verified controls stop. It reads your usage data, controls and evidence, writes down its reasoning and keeps the inputs with the result.

When data is missing, the agent names the gap and does not produce a number.

Read the white paper on Fermi estimates

Input manipulation · Claims triage assistantMonday 24 Aug, 09:00

0.50incidents a year

€400,000loss per incident

52%not stopped by verified controls

€104,000a year

Risk Agent reasoning

The input validation test failed on the new channel. Mitigation factor lowered from 0.55 to 0.48.

UsageMonthly active users and live channels from the connected source
ControlsRate limiting evidenced · input validation test failed on the new channel
LossClaims handling cost and regulatory exposure per incident

RT-189 Input manipulation · risk value over timeNext run Mon 28 Sep, 09:00

Mon 24 AugInput validation test failed on the new channel€104,000+14,000
Mon 10 AugSecond customer channel live€90,000+18,000
Mon 13 JulMonthly active users up 15%€72,000+9,000
Mon 15 JunInput validation control passed its runtime test€63,000−51,800
Mon 1 JunRun failed, usage source did not respond, previous value kept€114,800
Mon 25 MayRate limiting on the public endpoint evidenced€114,800−11,200
Mon 4 MayFirst run, no verified defence against crafted inputs€126,000

21 weekly runs since 4 May: 20 quantified, 1 failed. The 14 runs with no change are not listed. Illustrative data.

The Risk Agent re-prices every risk on your schedule

Choose a cadence once: daily, weekly, monthly or quarterly. Every run is dated and kept, so the history builds up without anyone opening a spreadsheet. If a source does not respond, the run is marked as failed, the last value stays in force and the gap is on record.

See what each control is worth

When a control passes its runtime test, the Risk Agent raises the mitigation factor and the estimate falls. When the test fails, the estimate rises again at the next run. Security spend gets a figure that a CFO can set against its cost.

Input validation control · Input manipulation on the Claims triage assistant

15 JunControl passed its runtime testShare not stopped 82% → 45%−51,800 a year
24 AugRuntime test failed on the new channelShare not stopped 45% → 52%+14,000 a year

Worth at today’s usage€60,000 a year

How controls and runtime tests work

Report exposure against the limits your board approved

Set a Risk Limit for the organization and split it across business units and AI systems, evenly or in proportion to economic value. After each run the roll-up shows where exposure stands against every limit, so the board pack starts from current figures.

€143.8M of the €159.5M organization limit

90% used

Monday 21 Sep, 09:00

Claims111%
Customer service91%
Finance83%
Sales77%
IT91%
Underwriting93%
Marketing105%
Legal and compliance83%
Operations77%
HR87%
Procurement113%
Actuarial80%

Exposure Risk Limit

Your risk team keeps its own models

Some risks deserve a model built by hand. Four other methods sit beside the Risk Agent, and all of them write into the same history and the same limits. Each estimate keeps its method, inputs and reasoning, so a reviewer can challenge any figure.

Monte Carlo simulation: Frequency and severity distributions from calibrated estimates.
Scenario analysis: Structured assumptions with ranges and transparent logic.
Risk matrix: For registers that still report likelihood and impact.
Manual entry: A figure your team has already defended elsewhere.

Bring one AI system to the demo

Tell us about one AI system you run. We walk through how the Risk Agent would estimate its main threats, and how schedules and limits would work across all your AI systems.