Sharing data for research, when carried out responsibly, can have huge public benefits. However, without appropriate protections in place, institutions risk losing the trust of individuals. Hence, privacy risk assessment should be baked into the decision-making processes for sharing or providing access to data. The current approaches for assessing privacy risk are ad hoc, manual, opaque, and inconsistent across different organisations or even different individuals in the same organisation. In this report, we propose a new privacy risk assessment framework that can improve consistency and transparency in data sharing decisions. Our intention is to support shared subjectivity in decision-making among various stakeholders and enforce the subjective decisions consistently. Our privacy risk assessment framework is built on top of the Five Safes, which is widely used across different public institutions in the UK. In the first PRiAM report (D1), we explored how various organisations using the Five Safes framework interpret it differently. It is impossible to assess if the framework is being used effectively, unless more details regarding how each of these safes were accounted for are available. The proposed privacy risk assessment framework aims to facilitate better usage of the Five Safes. The key idea is to enable data custodians to explicitly list the criteria they consider for assessing privacy risk, thereby enhancing transparency. These criteria are then used to categorise different data sharing scenarios into discrete tiers of risk that can further be tied to decisions around data sharing, therefore providing consistency in decision-making. Creating discrete levels of risk encourages comparison-based reasoning about risk in different scenarios as well as provides a starting point for the creation of standard benchmarks.