Open access to AI foundational models poses various security and compliance risks, report finds

The report from the Institute for Security and Technology assesses security and compliance risks across a spectrum of AI foundation model accessibility, finding that risks such as malicious use and accountability gaps increase as models become more open. It emphasizes that fully open releases are irreversible, advocates for gating mechanisms to provide traceability, and underscores that AI can support both offensive misuse and defensive cybersecurity efforts.