Skip to content

Misuse Risks

Misuse risks arise when humans deliberately use AI systems to cause harm. Unlike accident risks, these require malicious intent—but AI dramatically amplifies what bad actors can do.

steals researchenables hackingprovides contentimpersonationjustifiesCyberweaponsBioweaponsAutonomousWeaponsDeepfakesDisinformationAI PoweredFraudMassSurveillance
increases

AI enabling new forms of violence:

RiskDescription
BioweaponsAI-assisted pathogen design and synthesis
CyberweaponsAutonomous hacking and vulnerability exploitation
Autonomous WeaponsLethal autonomous weapons systems (LAWS)

AI-powered influence and fraud:

RiskDescription
DisinformationAI-generated propaganda and influence operations
DeepfakesSynthetic media for impersonation and fabrication
AI-Powered FraudAutomated scams, social engineering, impersonation
RiskDescription
Mass SurveillanceAI-enabled monitoring at scale

Surveillance and other misuse capabilities can enable Authoritarian Takeover (a structural risk). The tools and techniques that enable authoritarian control are documented in Authoritarian Tools.


AI lowers barriers to sophisticated attacks. A lone actor can now generate personalized disinformation at scale, or potentially access dangerous knowledge that previously required rare expertise.

Many misuse risks shift the balance toward offense. It’s easier to generate disinformation than to detect it, easier to find vulnerabilities than to patch them.

Most AI capabilities have both beneficial and harmful applications. Language models that help with coding can also help with malware. Models that accelerate drug discovery could accelerate bioweapon design.


  • Misuse is more dangerous when AI systems are more capable
  • Accident risks become more severe if misused AI is harder to control
  • Both require understanding AI capabilities we don’t fully control

These misuse risks are influenced by the following amplifiers from other risk categories:

FactorHow It Contributes
Authoritarian ToolsCapabilities that enable surveillance and control
ProliferationMore actors gain access to dangerous capabilities
Authentication CollapseHarder to distinguish real from fake content