A trust policy is a cloud rule that says which identities are allowed to take on a given role and use its permissions. It is separate from the permissions the role carries: one defines who may use the role, and the other defines what the role can do.
It matters because a role that is too easy to take on is a way into an account. If the trust policy accepts a broad range of identities, for example any workflow from any repository in an organization, then anyone who controls one of them can use the role.
Details differ between cloud providers, including the names used and the conditions that can be set. A policy that works is not necessarily a safe one, and changes to it can be easy to overlook in reviews because it looks like a short configuration rather than a permission.
To keep it safe, name the specific identities, repositories and branches that may use a role, avoid wildcards, review changes to trust policies as carefully as changes to permissions and test that other identities are refused.