Among the potentially hazardous asteroids (PHAs) in orbit around the Earth, some of them can become a real threat. The most famous PHA presently known is asteroid (99942) Apophis which briefly presented an unusually high impact probability (up to 2.3 %) for a collision with the Earth in 2029. It remains the only asteroid to have reached level 4 of the Torino Scale. Even if Apophis is not a threat anymore, other PHAs are still monitored and now, only one asteroid is scaled to 1 with the highest impact probability. Asteroid 2011 AG5 has 1 chance over 500 to hit the Earth on 2040. This asteroid is challenging because it will remain of faint magnitude around 23.0 until its close encounter with the Earth in February 2023. It will come close to the Earth by 0.012 AU. Intensive ground-based (optical and mainly radar measurements) will be performed. Before this date, optical measurements would be possible (provided that large telescopes are used) and orbital refinement could be performed in order to improve the orbital uncertainty of this asteroid. Nevertheless, no physical data can be derived before 2023 and therefore, the influence of non gravitational forces, mainly Yarkovsky effect, can not be precisely determined. This non gravitational effect produces a secular drift da/dt (positive or negative) of the semi-major axis due to the anisotropic re-emission of the incident solar radiation. We propose here a dynamical study of the asteroid 2011 AG5. We discuss first the location of primary and secondary keyholes in the target plane of 2023 as well as the quantification of the impact probability. Secondary keyholes are due to two consecutive close encounters, the second usually happening near a keyhole or a resonant return. Then, we will address how those quantities evolve with future dedicated ground-based measurements. In a second part, we will discuss non gravitational perturbations through Yarkovsky effect. Assuming that this asteroid is a C or S-type, we can statistically derive some maximum intensity of Yarkovsky force, without any assumptions on the physical parameters. This will help to assess the maximum deviation expected on the geocentric distance expressed in the 2023 target plane. This deviation will have a direct consequence on the impact probability. Finally, a deeper study will include a Monte Carlo test on the orbital fit in order to compute virtual asteroids (VA) moving under gravity, relativistic and Yarkovsky perturbations. Using a simple model of Yarkovsky force as a perturbation along the transverse component and inversely proportional to the heliocentric square distance of the asteroid, we include a random deviation da/dt to assess the number of VA becoming virtual impactors (VI). We will compare this number to the one obtained with VA moving only under gravity and relativistic perturbations.