Back to the MIT repository
4. Malicious Actors & Misuse2 - Post-deployment

Malicious Uses

As AI assistants become more general purpose, sophisticated and capable, they create new opportunities in a variety of fields such as education, science and healthcare. Yet the rapid speed of progress has made it difficult to adequately prepare for, or even understand, how this technology can potentially be misused. Indeed, advanced AI assistants may transform existing threats or create new classes of threats altogether

Source: MIT AI Risk Repositorymit376

ENTITY

1 - Human

INTENT

1 - Intentional

TIMING

2 - Post-deployment

Risk ID

mit376

Domain lineage

4. Malicious Actors & Misuse

223 mapped risks

4.0 > Malicious use

Mitigation strategy

1. Mandate and conduct systematic *adversarial testing* and *red teaming* throughout the model lifecycle to proactively identify vulnerabilities and assess differential risk, thereby establishing adversarial resilience as a foundational prerequisite for deployment. 2. Implement a multi-layered defense architecture utilizing robust *input/output validation* and *sanitization* mechanisms, supplemented by real-time *anomaly detection* and protective models (e.g., Prompt Shields) to prevent exploitation via techniques such as prompt injection and evasion attacks. 3. Establish comprehensive *AI Security Compliance* programs, requiring stakeholders to adopt and adhere to established best practices, IT reforms, and detailed response plans to mitigate the systemic risk and lower the impact of successful malicious use.