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

Malicious Use (Intentional)

empowering malicious actors to cause widespread harm

Source: MIT AI Risk Repositorymit340

ENTITY

1 - Human

INTENT

1 - Intentional

TIMING

2 - Post-deployment

Risk ID

mit340

Domain lineage

4. Malicious Actors & Misuse

223 mapped risks

4.0 > Malicious use

Mitigation strategy

1. Implement rigorous biosecurity protocols, including strict access controls and increased scrutiny for AI systems with biological research or engineering capabilities, to prevent their repurposing for bioterrorism. 2. Restrict access to dangerous AI models and computational resources by employing mechanisms such as structured access via controlled cloud services, mandatory know-your-customer screenings, and hardware/export controls on high-risk capabilities. 3. Employ advanced model-level misuse prevention techniques, such as fine-tuning to condition models to refuse explicit harmful requests, deploying always-on detection and filtering systems for both user inputs and model outputs, and integrating safety-enforcing system prompts.

ADDITIONAL EVIDENCE

In this section, we will explore the various ways in which the malicious use of advanced AIs could pose catastrophic risks. These include engineering biochemical weapons, unleashing rogue AIs, using persuasive AIs to spread propaganda and erode consensus reality, and leveraging censorship and mass surveillance to irreversibly concentrate power.