glossary · the vocabulary
the language ofinsuringai.
the coverage
terms, defined
ai agent insurance
liability coverage for the harm an autonomous software agent can do while acting on your behalf: wrongful refunds, invented policies, leaked data, deleted records, injected behavior. goodfault writes it as eight named perils, a1 to a8, affirmatively and by name. also called agentic ai insurance or ai liability insurance.
agentic ai insurance
another name for ai agent insurance: cover for ai systems that plan and act with autonomy, not just answer questions. the exposure is what the agent does, not what it says.
ai liability insurance
the broad category of cover that responds when an ai system causes financial loss, bodily injury, or property damage. goodfault writes the ai-specific perils affirmatively rather than leaving them to silent coverage.
physical ai insurance
cover for ai that acts in the physical world: humanoids, autonomous mobile robots, drones, ai-embedded products, and the software layer of autonomous vehicles. when ai gets a body, its errors become bodily injury and property damage, not only financial loss. goodfault writes it as division b, perils b1 to b5.
humanoid robot insurance
liability cover for humanoid and general-purpose robots working near people in industrial, logistics, hospitality, and domestic settings. bodily injury, third-party property damage, and the operational perils of machines built to move among us. goodfault writes it as peril b1.
drone insurance
commercial drone liability and hull cover, written for ai autonomy and priced per mission profile, usage-based where operations justify it. goodfault writes it as peril b3.
named peril
a risk written into the policy explicitly, by name, so it is covered on purpose. the opposite of silent coverage: at claim time the question is which peril fired, not whether ai was quietly excluded.
affirmative ai coverage
cover that names ai risk and grants it, rather than staying silent or excluding it. goodfault's whole book is affirmative, perils a1 to b5, written by name.
silent ai
ai risk that a policy neither names nor excludes, the ai analogue of silent cyber. it leaves both insurer and insured exposed to an argument at claim time. named perils remove the ambiguity.
how it is priced
terms, defined
blast radius
how much damage a system can cause before a human can stop it: how many customers, records, systems, or square meters one bad decision touches. the single best predictor of a loss we have found.
authority ceiling
the largest action a system can take unattended: the biggest transaction an agent can execute, or the force, speed, and workspace a machine can operate in near people. lower the ceiling, lower the premium.
autonomy tier
how much a system does without a human in the loop, from human-approves-everything to full autonomy with sub-agents. a core pricing factor, because autonomy is what turns a mistake into a loss.
control credits
premium discounts earned by safeguards enforced in code: caps, circuit breakers, rollback, kill-switches, geofencing, plus a governance credit up to 20% for passing recognized certification. controls that work make premiums fall.
parametric insurance
cover that pays on a defined trigger instead of a negotiated adjustment. when the parameter fires, payment follows, so example scenarios settle in days, not quarters.
audit-gated underwriting
binding cover only for systems whose record can be trusted. hold a recognized audit and you are in scope, quoted in minutes; no audit yet and goodfault refers you to independent auditors, then binds once you are certified.
carrier, not auditor
goodfault's neutral stance: it recognizes independent third-party audits and does not run its own, so the certificate stays meaningful to the enterprise buyers and regulators who rely on it.
the market
terms, defined
cg 40 47
the iso general-liability endorsement family, the 01 26 edition and related forms, that lets insurers exclude bodily injury, property damage, and advertising injury arising out of generative and other ai. it began hitting us renewals on january 1, 2026, and is the gap goodfault writes into affirmatively.
ai e&o
errors-and-omissions exposure created by ai: an agent gives wrong advice, invents a commitment, or takes an incorrect action. traditional e&o assumes a human operator, so many wordings now exclude or ambiguously handle autonomous ai. named perils close that gap.
eu ai act
the eu regulation whose high-risk obligations require deployers to show human oversight, retained logs, and a named accountable human, with fundamental-rights impact assessments where required. peril a7 covers regulatory investigation and defense cost, and the evidence the act expects is the same evidence that earns control credits.
the vocabulary isthe coverage.
every term here maps to something on our paper. see the named perils on coverage, how they are priced on how it works, or the direct answers on the faq.
like what you see?
get covered.
tell us what your agent does, what tools it can reach, and what a bad day looks like. we return a premium in plain english. no actuarial latin, no 40-page pdf. most policies bind in under a week.
get a quoteconsult the oracle
not ready for a quote? describe your agent to the oracle and get an honest read of your riskiest failure mode. the one your roadmap is politely ignoring. free, anonymous, occasionally prophetic.
