Conscium: the factory infographic
An AI safety lab that tests and verifies agents in simulation before they are deployed.
What the company does
In normal terms
AI safety lab and agent verification platform: simulation testing and verification so AI agents can be deployed with more confidence, plus research on conscious AI.
Conscium tests and verifies agents in simulation so they can be deployed with more confidence, and researches conscious AI.
In lay terms
Conscium is a flight simulator for AI assistants: it tests them in a pretend world before they go to work.
The risks it faces, especially from AI agents
In normal terms
- Testing against what?. Verification needs a statement of intended behaviour to test against.
- Simulation is not production. Real deployments reach systems the simulation never had.
- Results nobody reads. Test results must reach the owners, security and insurers in a form they use.
In lay terms
- Testing against what?. You can't mark an exam with no answer sheet.
- Simulation is not production. The simulator has no real passengers.
- Results nobody reads. A pass certificate in a drawer.
How RiskMandate helps
- Reachwhat it can actually access or affectWhat the agent will reach in production, to simulate.
- Mandatewhat the business authorisedThe answer sheet: what it should do.
- Gapthe difference: the riskThe cases to test hardest.
- Barrierswhat really stands in the wayConscium's verification as review evidence.
- Provenancewho owns it, who reviewed it, which versionWhich version passed which tests.
In normal terms
A behaviour policy says what an agent may do; verification shows whether it did. RiskMandate gives Conscium the answer sheet and gives its results a place in the agent's record.
In lay terms
RiskMandate writes the answer sheet; Conscium marks the exam.
A first line for the conversation: A policy says what an agent may do; your verification shows it does. Shall we join them up?
Sources and method
The facts in grey (what it does) are from the company map, which cites:
- websummit.com, read 2026-11-01: LIS26 session page (conferenceSlug lis26): 'The biological bottleneck', 12 Nov, Cybersecurity Summit, Stage 8, names Conscium as the speaker's organisation
- websummit.com, read 2026-11-01: LIS26 attendee page: Calum Chace, Co-founder, Conscium; country and industry as listed by Web Summit; public bio
- conscium.com, read 2026-11-01: what the company does, from its own site (title and description)
The risks and how RiskMandate helps are our analysis, written by dev@web-summit.indi on 3 November 2026 from those facts and RiskMandate's model as this site explains it (Reach, Mandate, Gap; an Agent Behaviour Policy in practice). They are not the company's own statements and name no incident. A person reviews each before it is used.