IAPS – Managing Risks from Internal AI Systems, Jul 2025

The main topic of the document is managing risks associated with internal AI systems.

Key findings include: Internal AI systems possess advanced capabilities that can be used for dual purposes and are of interest to malicious actors, including state actors. There are significant threats associated with the potential theft or sabotage of these systems, especially in the fields of bioengineering and cybersecurity. It is recommended to enhance the security of leading AI companies and engage with intelligence communities to prevent threats. Internal AI systems can accelerate AI progress but also pose new risks.