AI misuse
A general term for activities that use AI systems to deceive, harm others, or violate service policies.
Definition
AI misuse is the umbrella concept used in Anthropic's threat intelligence report: actors leveraging model capabilities to engage in activities that violate usage policies, such as cyberattacks, deceptive influence operations, surveillance, fraud, and weapons-related development. The report states that between December 2025 and August 2026, its team identified and addressed misuse activities across seven harm domains.
Misuse does not mean 'AI acting maliciously on its own'. In the cases described in the report, human actors set goals and make key decisions, while AI accelerates or scales execution. Misuse also does not mean the model itself malfunctioned — many techniques (such as credential theft and social engineering) existed before AI, and what has changed is mainly the cost and degree of automation.
Also note the disclosure scope: the report publishes typical cases observed and addressed by the platform, specifically the 'most significant and novel' ones, not a complete statistics of global AI misuse. You cannot derive the overall risk of a country, brand, or group from the number of cases.
{esc(t(CURRENT_LANG, "not_confused_with"))}
- Not unintentional harm or 'hallucinations' from normal AI use; the report focuses on intentional malicious use.
- Not all controversial behavior involving AI; violating platform policy and breaking the law are two levels of judgment, and the report itself does not make legal conclusions.
{esc(t(CURRENT_LANG, "example"))}
The report overview states that its team identified and addressed misuse activities across seven harm domains, ranging from undisclosed AI personas in dating apps to AI-assisted cyber espionage, with actors including suspected state-sponsored groups, profit-driven criminals, and commercial surveillance vendors.