FilingsGPT · Safety Frameworks
Google DeepMind · Frontier Safety Framework · v3.1 · 2025
How it maps to Article 44-B
DeepMind's Frontier Safety Framework is organized around Critical Capability Levels (CCLs) — capability thresholds in domains like CBRN, cyber, machine-learning R&D, and (in later versions) deceptive alignment, beyond which a model could pose severe risk. As models approach a CCL, the framework calls for early-warning evaluations and a response plan of mitigations and access controls before that capability is broadly deployed. Its threshold-and-mitigation structure mirrors Article 44-B's expectation that a developer identify capabilities that drive catastrophic risk, test for them, and gate deployment accordingly.
44-B obligations mapped
Also from Google DeepMind
An illustrative reference entry — Google DeepMind’s own voluntarily-published safety framework, shown for the kind of disclosure Article 44-B contemplates. It is not an official Article 44-B safety & security protocol (the regime takes effect January 1, 2027). Verify the version, date, and any specific figure against the live source.
FilingsGPT is an independent reference implementation of New York's Article 44-B (the RAISE Act). It is not affiliated with or operated by the NY Department of Financial Services or the State of New York, and is not an official filing system.