https://coursiv.io/blog/tags/trust-risk--safetyRecent content in Trust, Risk & Safety on Coursiv BlogHugo -- 0.147.0en-UShttps://coursiv.io/blog/glossary/adversarial-exampleMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/adversarial-exampleAn input perturbed just enough to fool a model, while looking unchanged to a person.https://coursiv.io/blog/glossary/ai-alignmentMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/ai-alignmentThe problem of making AI systems pursue what we actually want, including things we never thought to specify.https://coursiv.io/blog/glossary/ai-incidentMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/ai-incidentAn event where an AI system causes or nearly causes harm — increasingly something you must record and report.https://coursiv.io/blog/glossary/ai-safetyMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/ai-safetyThe field concerned with preventing AI systems from causing harm, from everyday failures to systemic risks.https://coursiv.io/blog/glossary/algorithmic-biasMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/algorithmic-biasWhen a system produces systematically different outcomes for different groups, without justification.https://coursiv.io/blog/glossary/benchmarkMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/benchmarkA standard test set used to compare models — useful for direction, unreliable as a guarantee.https://coursiv.io/blog/glossary/biasMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/biasSystematic skew in a model's outputs — a data and design property, not a moral accusation.https://coursiv.io/blog/glossary/content-credentialsMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/content-credentialsCryptographically signed metadata recording how a piece of media was created and edited.https://coursiv.io/blog/glossary/data-poisoningMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/data-poisoningCorrupting training data so the resulting model misbehaves, often only on a specific trigger.https://coursiv.io/blog/glossary/data-provenanceMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/data-provenanceA documented record of where data came from, how it was collected, and what may be done with it.https://coursiv.io/blog/glossary/differential-privacyMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/differential-privacyA mathematical guarantee that a result barely changes whether or not any one person's data was included.https://coursiv.io/blog/glossary/evaluationMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/evaluationSystematic measurement of whether an AI system does what you need — your tests, not the vendor's.https://coursiv.io/blog/glossary/explainabilityMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/explainabilityBeing able to give a human a meaningful account of why a system produced a particular output.https://coursiv.io/blog/glossary/fairnessMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/fairnessThe requirement that a system treat people equitably — with several mathematical definitions that provably conflict.https://coursiv.io/blog/glossary/guardrailsMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/guardrailsControls that constrain what a model can be asked or allowed to output.https://coursiv.io/blog/glossary/interpretabilityMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/interpretabilityUnderstanding how a model works internally, as opposed to just describing its outputs.https://coursiv.io/blog/glossary/jailbreakMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/jailbreakA prompt crafted to get a model to bypass its own safety rules.https://coursiv.io/blog/glossary/model-cardMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/model-cardA short standard document stating what a model is for, how it was evaluated, and where it fails.https://coursiv.io/blog/glossary/model-driftMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/model-driftSilent degradation as the world moves away from the data a model was trained on.https://coursiv.io/blog/glossary/prompt-injectionMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/prompt-injectionHiding instructions in content the model reads, so a third party effectively takes over the prompt.https://coursiv.io/blog/glossary/red-teamingMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/red-teamingDeliberately attacking your own AI system to find failures before someone else does.https://coursiv.io/blog/glossary/responsible-aiMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/responsible-aiThe practice of building and deploying AI with its effects on people deliberately considered.https://coursiv.io/blog/glossary/robustnessMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/robustnessWhether a system keeps performing when inputs are noisy, unusual, or deliberately adversarial.https://coursiv.io/blog/glossary/shadow-aiMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/shadow-aiEmployees using AI tools that the organisation has not approved and cannot see.https://coursiv.io/blog/glossary/transparencyMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/transparencyMaking it clear what an AI system is, what it was trained on, and when someone is interacting with it.https://coursiv.io/blog/glossary/trustworthy-aiMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/trustworthy-aiAn umbrella for the properties an AI system needs to be relied on: valid, safe, secure, accountable, explainable, fair, privacy-enhanced.https://coursiv.io/blog/glossary/watermarkingMon, 01 Jan 0001 00:00:00 +0000https://coursiv.io/blog/glossary/watermarkingEmbedding a detectable signal in AI-generated content to mark its origin.