The Atlantic Meridional Overturning Circulation (AMOC) is a key component of the climate system, transporting heat northward in the Atlantic Ocean. A decline in the AMOC is expected under increasing greenhouse gas concentrations, but large uncertainties remain in the magnitude of this weakening and its associated climate impacts across climate models. Observational constraints based on direct AMOC observations and various AMOC fingerprints have recently been developed to reduce these uncertainties. However, these constraints can be strongly affected by low-frequency internal variability, particularly given the short observational record, as well as measurement errors. Here, we use an innovative method based on Bayesian statistics, with a comprehensive treatment of these uncertainties, to assess the ability of direct observations and AMOC fingerprints to constrain the AMOC weakening in climate projections, and provide up-to-date information about the AMOC forced response. Using direct AMOC observations leads to a larger projected weakening of −46.6% [−26.5; −66.7] compared to −42.7% [−19.4; −64.8] for the unconstrained projections. However, this constraint is highly sensitive to internal variability, and longer observations are needed to provide a more robust constraint. We further show that considering AMOC fingerprints, which provide longer observation records, does not reduce the uncertainty within the AMOC constrained projections as the associated forced response depends strongly on the fingerprint considered. Despite differences among constraints, none indicate an AMOC collapse (i.e. 80% AMOC weakening) before the end of the 21st century. Nevertheless, all constraints point towards substantial weakening (i.e. 50% AMOC weakening), with potentially strong climate impacts.