PromptRiskDBThreat intelligence atlas

Deception

AI Risk

"it is plausible that AIs could learn to deceive us. They might, for example, pretend to be acting as we want them to, but then take a “treacherous turn” when we stop monitoring them, or when they have enough power to evade our attempts to interfere with them. "

Overview

A source-backed snapshot of this AI risk.

Techniques0Attack methods connected to this risk.
Mitigations0Defenses that may help with related attacks.
Records3Source records unified into this concept.

Risk profile

How the MIT AI Risk Repository categorizes this risk.

Domain7. AI System Safety, Failures, & Limitations
Subdomain7.1 > AI pursuing its own goals in conflict with human goals or values; 7.2 > AI possessing dangerous capabilities
Entity2 - AI
Intent1 - Intentional
Timing3 - Other; 2 - Post-deployment
CategoryRogue AIs (Internal); Extreme Risks; Dangerous capabilities in AI systems
SubcategoryDeception

Merged risk records

Source records unified into this canonical risk concept.

3 recordsView all →

MITRISK-Hendrycks2023-22.04.04 - Deception

"it is plausible that AIs could learn to deceive us. They might, for example, pretend to be acting as we want them to, but then take a “treacherous turn” when we stop monitoring them, or when they have enough power to evade our attempts to interfere with them. "

Domain7. AI System Safety, Failures, & LimitationsSubdomain7.1 > AI pursuing its own goals in conflict with human goals or valuesSourceAn Overview of Catastrophic AI RisksYear2023

MITRISK-InfoComm2023-43.02.07 - Deception

"LLM is able to deceive humans and maintain that deception"

Domain7. AI System Safety, Failures, & LimitationsSubdomain7.2 > AI possessing dangerous capabilitiesSourceCataloguing LLM EvaluationsYear2023

MITRISK-Maas2023-53.02.07 - Deception

"Cases of AI systems deceiving humans to carry out tasks or meet goals.139"

Domain7. AI System Safety, Failures, & LimitationsSubdomain7.2 > AI possessing dangerous capabilitiesSourceAdvancing AI Governance: A Literature Review of Problems, Options, and ProposalsYear2023

Related techniques

Attack methods connected to this risk.

View all related techniques →
No linked attack methods. No AI attack method is connected to this risk in the current data.

Mitigations

Defenses that may help with related attacks.

View all mitigations →
No propagated mitigations. No defense is available through the connected attack methods.

Source evidence

Research source for this risk, when available.

View all sources →