One challenge with using carbon offsets to compensate for CO2 emissions is that the effects of CO2 emissions are very long-lived. Most of the carbon in a tonne of CO2 emitted today will – eventually – be removed from the atmosphere. However, around 25% remains in the atmosphere for hundreds to thousands of years. To compensate for this, offset credits must be associated with GHG reductions that are similarly permanent. If a GHG reduction or removal is “reversed” (i.e., GHGs are subsequently emitted so that no net reduction occurs), then it no longer serves a compensatory function.
For most kinds of carbon offset projects, reversals are either physically impossible or extremely unlikely. The greatest risk occurs with projects that store carbon in “leaky” reservoirs. The classic example is a forestry project that keeps carbon in trees and soils (and adds to those carbon stores over time, as the forest grows). Such a project will reduce CO2 emissions – and increase removals – if the trees would have been cut down otherwise. But, if a fire later burns down the project’s trees, some or all of the carbon may be (re)emitted, leading to a reversal.
One common misunderstanding is that – for carbon offsets – “permanent” means something less than hundreds or thousands of years. A standard convention, for example, is that carbon only needs to be kept out of the atmosphere for 100 years (or less, in some cases) to be considered “permanent.” Such compromises are frequently made in the context of carbon offset programs seeking to balance technical requirements with the practical constraints of insuring against reversals. But, scientifically, anything less than a full guarantee against reversals into the indefinite future is not “permanent.”
 Technically, the individual molecules of CO2 emitted may cycle back and forth between the atmosphere and terrestrial reservoirs multiple times, but atmospheric concentrations of CO2 will remain elevated by an amount equal to about 25% of the original mass emitted after 1,000 years (Joos et al. 2013).
 Technically, a “reversal” occurs if – at any point in the future – the rate of GHG emissions accelerates so that it is higher than would have occurred if the project had never happened. For example, protecting a parcel of forest from deforestation prevents 100 tonnes of carbon from being released to the atmosphere (reducing emissions by 100 tonnes). Fifty years later, however, the parcel is burned down, emitting all the carbon. The rate of emissions in year 50 is accelerated, because without the project, the 100 tonnes of carbon would not have been present to be burned. Net GHG reductions over 50 years are zero, because the additional emissions cancel out the prior reductions.
 Unless a project involves carbon storage of some kind (e.g., sequestering carbon in trees), a reversal of emission reductions is highly unlikely. In theory, however, reversals can occur in other circumstances. A hypothetical example would be where a solar panel and battery storage system is used to provide electricity to a building, allowing it to operate off the grid (and avoiding grid-based electricity emissions); however, the solar panel fails and a backup diesel generator is brought in to provide power instead, causing more emissions than would have occurred without the project. Such circumstances will be rare, but for some project types, it may be worth evaluating whether these types of risks exist.