An autonomous system is a set of routers and addresses, under one or more administrations, which presents a common routing policy to the Internet  An older definition, that it was a set of routers or addresses under a common administration, is obsolete; the term routing domain is preferred for the set under single administrative control. This older definition did not cover the common case of a group of enterprise AS talking to an ISP AS, and that the ISP's policy is the only one seen by the rest of the Internet.
It is entirely possible that a single Internet AS actually includes multiple organizations, each with different policies on what can and cannot be sent to the Internet. If, however, they all go through the same Internet Service Provider (ISP), it is the rules of that ISP that override all others. In other words, a given customer of an ISP can choose to restrict types of Internet access (e.g., certain types of peer-to-peer file sharing protocols) or destinations (e.g., web servers offering online gambling). If the ISP, however, also blocked a voice over internet protocol service or a class of political content servers, those rules are in addition to those of the customers.
Autonomous system administration
Each AS has an autonomous system number (ASN), currently 16 bit but being expanded to a 32 bit space. ASNs are assigned by address registries.
Justification for AS assignment
Policies differ among registries, but, assuming the AS requester uses IPv4 address space and routes it on the public Internet, the minimal justification is to have a /20 or larger address block, or to have a address block of /24 or larger, which is multihomed to at least two other AS.
Private AS space
In the 16-bit AS space, the ASNs 64512 through 65535 are reserved and will never be routable on the public Internet.
Introduction to AS policies
There are many possible relationships among AS. Among the most basic is an economic one, in which the AS consider one another as equals, in which each believes that their customer base is approximately the same size, and it is of mutual benefits that their customers be able to reach one another. In such cases of peering,. When the parties agree that they are peers, they advertise, to each peer, the address space (i.e., of their internals and of their customers) to which they offer connectivity. No money changes hands because this is considered a balanced exchange.
In contrast, a transit relationship involves the transfer of money from the customer AS to the provider AS; the transit provider accepts money for providing connectivity to all reachable Internet destinations. See telecommunications provider economics for a more detailed discussion of some of the models used both in traditional telephony and in the Internet.
Many other relationships apply. Certain high-bandwidth academic and research AS, just as did the original ARPANET, may only offer access, or perhaps also transit, to members of their academic and research community. It is not coincidental that BGP can identify groups of addresses with a tag called a BGP community.
Some AS are, for various reasons, untrusted. A given AS may not accept a BGP advertisement that includes a route that passes through such an AS.
- Hawkinson, J. & T. Bates (July 1996), RFC 1930: Guidelines for creation, selection, and registration of an Autonomous System (AS)
- "Peering", unfortunately, is also used as a term for the two parties in a Border Gateway Protocol session. While AS do interconnect using BGP, this discussion is at a level higher than individual connections
- Norton, William B. (November 2003), "The Evolution of the U.S. Internet Peering Ecosystem", North American Network Operators Group