Meeting net-zero targets requires scaling carbon dioxide removal from ∼2 GtCO2/yr today to 7–9 GtCO2/yr by 2050. Since different carbon removal technologies face binding upper limits on deployable scale, this will require trillions of dollars of investment to be allocated across technologies with fundamentally different risk and cost profiles. Yet carbon removal markets lack a standardised, quantitative measure of permanence risk. Inspired by Value at Risk in financial markets, we propose Carbon at Risk (CaR): a probabilistic measure of the shortfall between contracted removals and realised storage at a given confidence level and time horizon. We calibrate CaR in two proof-of-concept applications: forest carbon, where Monte Carlo simulations driven by satellite fire data yield a single-project 95% CaR of more than 70% over 200 years for California, and geological storage (DACCS), where 95% CaR ranges from 0.15% (well-regulated offshore) to 17% (poorly regulated onshore). This variation implies that current permanence mechanisms, typically flat-rate deductions applied once at credit issuance, systematically over-buffer low-risk projects while under-buffering high-risk ones, and could be structurally ill-suited to risks that compound and shift over decades. We show that combining technologies in a portfolio creates a trade-off between cost and risk, and that the minimum cost of meeting a permanence target depends on the correlation structure of key risk factors. CaR provides a basis for risk-calibrated protection mechanisms, from buffer design and insurance pricing to fund-based instruments sized to cover reversal risk, across policy-relevant time horizon. It also provides the basis for technology-agnostic project comparison, reserve construction, and for the international accounting frameworks needed to steward global carbon stocks and sinks as a common resource.