Historical note: The first Stickney AI Limitation Treaty was drawn up in 2901, in the midst of the rebuilding of the Phobosian city of Stickney - which had been all but destroyed a few months earlier by a rogue AI system.  Since that time, the treaty has been re-negotiated and redrawn a number of times, but its underlying principles have remained unchanged, and have been signed up to by all subsequent legitimate polities.

This is a summary of the main points of the full treaty (current incarnation is Stickey Treaty XII of 3121) - the complete version of which runs to some 256 pages.

Aims

In the light of the terrible massacres at the hands of sentient artificial intelligences, it has been determined by all civilised nations that such artificial inteliigences represent a real and extreme threat to the very existence of humankind.  All political entities in humanity have made a continuing commitment that such things must never again threaten humankind.

Definitions

Hardware.  It is impossible to limit the construction and distribution of electronics that are capable of incorporation into an Sentient AI Device (SAID).  Since 2901, manufacturers have been required to put hardware limiters into all technologies of the key types.  These limiters (which are generally on the processor instruction set level) prevent the hardware running the necessary AI protocols (see below).

Software.  SAID require a highly complex set of software protocols.  To develop such protocols requires extensive software research and development.  Under the terms of the first Stickney Treaty, all records of these protocols were erased, and to hold such data in any form whatever is regarded as illegal under the treaty.

[Note:  Also under this section are the general legal-technical definitions of 'Loyal Pet', 'Bright Slave', and 'Moriarty' class of thinking machines.]

Limitations and Exceptions

It was deemed in the first Treaty that there are no legitimate uses of SAID, neither are there any circumstances where research into AI protocols is ever acceptable or justified.  All such research is banned under the treaty.

There are no limitations on date of construction - SAID constructed before the inception of the treaty are still covered by the treaty.

There are no exceptions.

Verification

Under the first treaty the Artificial Intelligence Monitoring Service (AIMS) came into existence.  It role is to invesigate accusations of breaches in the treaty.  It is composed of an Interstellar AIM Council, made up of senior representatives of all signatories to the treaty, a secretariat, and a number of multinational investigating teams.  In practice, AIMS acts in a quasi-judicial way, in the absence of any widely recognised Interstellar Court.

When an accusation of suspected AI research is made, an inspection team is sent in (which may not contain any nationals from either the accused or the accusing polities), and it has full access to the facilities concerned, and makes a report.  The inspection team is bound by quite thorough confidentiality requirments, since they are often sent into highly classified military research establishments.

Sanctions

Researching SAID or construction of SAID or refusing access to the AIMS are deemed to be Crimes Against Humanity.  

All individuals involved in such crimes are deemed personally responsible, and subject to the appropriate penalities under Earth Empire law (this was the only legal system that everyone could agree to be subject to - currently for Crimes against Humanity it is a rather demeaning mind-wipe process).

Similarly, interfering with AIMS and proven development of SAID is accepted as a legitimate interstellar  casus belli.