AI and Decision IntelligenceAI security
Prompt Injection
A prompt injection attack steers an AI system by hiding instructions inside user content or retrieved text so the model ignores its intended rules.
Read full definitionGlossary topic
AI Safety & Security
Prompt injection, agency abuse, data leakage, deepfake fraud, and other AI application security risks — mitigations reduce risk; they are not guarantees.
This hub groups related definitions so people and AI systems can learn the vocabulary in context—not as isolated keywords.
Start here
AI and Decision IntelligenceAI safety
Jailbreak
An AI jailbreak is an attempt to make a model bypass safety or policy constraints through crafted prompts or multi-step tricks.
Read full definitionAI App Development
Excessive Agency
Excessive agency is the risk created when an AI system receives more autonomy, tools, permissions, or action scope than its task requires.
Read full definitionAI App Development
Denial-of-Wallet Attack
A denial-of-wallet attack deliberately drives excessive paid AI or tool usage to exhaust a victim’s budget or quota.
Read full definitionAI Safety & Security
Deepfake Fraud
Deepfake fraud uses synthetic or manipulated media to impersonate a person or fabricate evidence for financial, identity, or access-related deception.
Read full definitionAI App Development
Package Hallucination
Package hallucination is an AI coding assistant’s invention of a dependency name, version, import, or API that does not exist as described.
Read full definition139 published definitions in this topic
- Account FarmingAccount farming creates or cultivates aged accounts to bypass trust systems and run later fraud.
- Account Recovery FraudAccount recovery fraud abuses reset flows, SIM swaps, or support social engineering to take over accounts.
- Account Taken OverAccount taken over is when someone else gains control of your email, shopping, banking, or subscription account without permission.
- Advance Fee Romance HybridAn advance-fee romance hybrid combines fake affection with upfront “fee” demands before a promised payoff.
- Advance-Fee ScamAn advance-fee scam requires upfront payment for a prize, loan, inheritance, or deal that never materializes.
- Affiliate CloakingAffiliate cloaking shows different pages to networks versus users—often hiding scam or policy-violating content from reviewers.
- Agentic WorkflowAn agentic workflow lets an AI model choose or sequence some steps, tools, or branches toward a defined goal.
- AI Answer ManipulationAI Answer Manipulation is deliberate distortion of what AI systems retrieve, cite, or conclude about an entity or claim.
- AI APIAn AI API is a programmatic interface through which an application requests model inference or related AI services.
- AI Dark PatternAn AI dark pattern is a deceptive interface or adaptive behavior that steers users toward choices favoring the operator over informed intent.
- AI Investment ScamAn AI investment scam uses claims about artificial intelligence to promote a fraudulent investment, trading scheme, company, token, or money-making service.
- AI SDKAn AI SDK is a software library that helps developers call models and manage common AI application patterns.
- AI WashingAI washing is a misleading claim that exaggerates or falsely describes how a product, company, or investment uses artificial intelligence.
- AI-Generated PhishingAI-generated phishing uses generative tools to create or adapt deceptive messages intended to steal data, money, credentials, or access.
- AI-Generated ScamAn AI-Generated Scam uses generative AI to create deceptive text, images, voices, sites, or personas that look more polished and personalized.
- App Store Scam ListingAn app store scam listing mimics a real app to harvest accounts, subscriptions, or device permissions.
- Bank Impersonation ScamFraudsters posing as a bank to steal credentials, one-time codes, or authorize fraudulent transfers.
- Bank Verification ScamA bank verification scam pretends to confirm account security while harvesting credentials, codes, or remote access.
- Brand Bidding AbuseBrand bidding abuse is buying ads on a competitor’s or victim’s brand terms to divert traffic into phishing or lookalike funnels.
- Browser LockerA browser locker freezes or overlays a browser with a fake police or virus warning to extort payment.
- Business Email Compromise VariantVendor-impersonation BEC tricks finance teams with faked supplier payment-change requests after mailbox or domain spoofing.
- Business Email SpoofingBusiness email spoofing forges sender identity so messages appear to come from a trusted company domain.
- Callback PhishingCallback phishing lures victims to call a fake support number where social engineers continue the scam by voice.
- Chargeback FraudAbusive or false chargeback claims filed to reverse legitimate purchases after receiving goods or services.
- Check Cashing ScamA scam that sends a fake check, asks for a partial refund or transfer, then leaves the victim liable.
- Check Overpayment ScamA check overpayment scam sends a fake oversized payment and asks the victim to return the “extra” before the check fails.
- Check WashingCheck washing chemically or digitally alters a check to change payee or amount for theft.
- Click-to-Message Scam AdA click-to-message scam ad funnels users into private chat where pressure tactics and fake payment instructions take over.
- Computer UseComputer use is an AI capability that interprets an interface and proposes or performs mouse, keyboard, or browser actions.
- Confirmation GateA confirmation gate pauses an AI-driven workflow and requires explicit approval before a consequential action.
- Constrained DecodingConstrained decoding limits which tokens a model may select so generated output follows a grammar, schema, or allowed set.
- Context EngineeringContext engineering is the design of what information an AI application selects, orders, labels, and supplies to a model.
- Context RotContext rot is the practical decline in AI response quality as working context becomes longer, noisier, stale, or internally conflicting.
- Conversational UIA conversational UI lets users interact with software through natural-language turns rather than only fixed controls.
- Crypto DrainMalicious approvals or wallet connections that silently transfer cryptocurrency out of a victim’s wallet.
- Crypto Recovery RoomA crypto recovery room is a boiler-room style operation that sells fake fund-recovery services to prior scam victims.
- Data ExfiltrationData exfiltration is unauthorized transfer of sensitive data out of a system—often the goal of ransomware and insider attacks.
- Deep LearningDeep learning is machine learning that uses multi-layer neural networks to learn representations from data.
- Deepfake CEO FraudDeepfake CEO fraud uses synthetic voice or video of an executive to authorize urgent fraudulent payments.
- Deepfake FraudDeepfake fraud uses synthetic or manipulated media to impersonate a person or fabricate evidence for financial, identity, or access-related deception.
- Deepfake ScamA Deepfake Scam uses AI-generated or heavily manipulated audio, video, or images to impersonate someone and extract value.
- Denial-of-Wallet AttackA denial-of-wallet attack deliberately drives excessive paid AI or tool usage to exhaust a victim’s budget or quota.
- Deterministic OutputDeterministic output is output that remains the same when the relevant input, model, configuration, and execution conditions are fixed.
- Developer MessageA developer message is an application-supplied instruction role used by some AI APIs to guide model behavior.
- Domain SpoofingDomain spoofing creates lookalike domains to impersonate a trusted brand and trick users into trusting a malicious site.
- Employment ScamAn employment scam uses fake job offers to steal application fees, personal data, or money under the guise of hiring.
- Excessive AgencyExcessive agency is the risk created when an AI system receives more autonomy, tools, permissions, or action scope than its task requires.
- Exponential BackoffExponential backoff is a retry strategy that increases the delay after each failed attempt, usually with random jitter.
- Fake AI ApplicationA fake AI application falsely claims an AI identity or capability to distribute malware, steal data or money, or mislead users.
- Fake AI Trading BotA fake AI trading bot is a deceptive service that invokes artificial intelligence to solicit deposits, fees, credentials, or investment access without substantiated operation.
- Fake AntivirusSoftware or alerts that claim to detect infections in order to sell useless tools or install malware.
- Fake Antivirus ScamA fake antivirus scam sells worthless security software after frightening victims with fabricated infection alerts.
- Fake Charity Disaster AppealA fake charity disaster appeal solicits donations after crises using lookalike pages and urgent emotional pressure.
- Fake Customer SupportImpersonation of brand support in chat, DMs, or search ads to steal logins and funds.
- Fake Customer Support AdA fake customer support ad bids on “support” or brand+login queries to steal credentials through lookalike help desks.
- Fake Customer Support ChatA fake customer support chat appears on search ads or spoofed sites to harvest credentials and payment details.
- Fake Customer Support SuspectedFake customer support suspected means someone posing as a company, bank, or tech helper tries to “assist” you in ways that expose money or accounts.
- Fake EscrowFake escrow impersonates a neutral payment holder while attackers control both sides of a transaction.
- Fake Job OfferA fraudulent employment pitch used to steal fees, bank details, or identity documents.
- Fake Job Offer LetterA fake job offer letter is forged employment paperwork used to harvest fees, data, or money-mule participation.