THE TOUGHEST challenge in framing a new strategic-arms agreement may be the problem of sea-launched cruise missiles, or SLCMs in arms-control jargon. At a time when the U.S. and Soviet navies are building large arsenals of non-nuclear SLCMs, how can we verify compliance with limits on nuclear ones?

Counting and verifying nuclear-armed SLCMs is difficult. The missiles are only about 20 inches in diameter and less than 30 feet long -- easily moved about, disguised and concealed. Furthermore, from the outside, a sea-launched cruise missile with a nuclear warhead looks identical to one which carries a conventional warhead. In a world containing thousands of SLCMS, constraining nuclear warheads to a small fraction of the missile force is a daunting task.

Schemes proposed for counting SLCMs involve either tagging nuclear missiles with a non-removable transmitter or placing monitoring devices on every ship, in both navies, that could launch a SLCM. Tagging in this fashion, however, is probably unacceptable to the two navies, because the tags must broadcast a signal that might give away the positions of ships at sea, possibly ships about to go to war.

In any case, tagging cruise missiles that are already nuclear-armed misses the real problem, which is preventing conversion of conventionally armed SLCMs to ones carrying nuclear explosives. That job must be done, moreover, withoutcompromising the operational security of the ships that carry the missiles and without intruding significantly on current operating procedures. Technology developed for other purposes provides precisely the tools necessary to meet these criteria.

American SLCMs, and perhaps Soviet ones as well, are currently mated to their warheads and placed in canisters before being placed in launch tubes or storage racks aboard ships or submarines. Normally, the canister is never opened unless -- and until -- the missile is fired. The canisters can be opened to service the missile and to change the type of warhead, although doing so at sea is not customary.

American nuclear weapons are currently protected from misuse and theft by devices called Permissive Action Links, or PALs. Before a weapon protected by a PAL can be detonated, a special code must be inserted into an electronic lock. If a mistake is made when the code is punched in, the PAL allows a small number of retries.

But a PAL goes much further in protecting a nuclear weapon. It can sense when an unauthorized attempt to get inside the bomb is being made; it can count the total number of wrong tries to insert firing codes; and if someone seeks to tamper with the weapon, the PAL can render the bomb and its finely machined nuclear components utterly useless. If the PAL receives the proper codes, it figuratively retreats from the picture and allows use of the weapon.

Just as PALs have guarded against unauthorized detonation of a nuclear warhead, they could seal a canister containing a conventional-payload SLCM and preclude the installation of a nuclear weapon. How might such a system work?

PALs usually consist of compact electronic locks using codes as long as 12 digits, although such codes could easily be extended to 24 or more digits. We could attach, for example, a PAL lock to any SLCM canister intended to contain a conventional warhead. The lock could require the insertion of a 24-digit code, half held by an American inspector, the other half by a Soviet counterpart. Such a lock would require, on average, literally thousands of years to "pick" by trying random combinations.

The side owning the missile would mate it with its warhead and place it in its canister. All of this could take place in full secrecy. Inspectors from the other country would then use passive radiation detectors to assure themselves that the canister contained no nuclear weapon -- it is virtually impossible to shield the launch canister of a nuclear payload well enough that the neutrons and gamma rays from its warhead would be undetectable. Finally, inspection teams from both sides would install the seals and set the combinations using random numerical sequences generated independently by each side. With the seals locked in place, the missile would be logged as a conventional SLCM.

A further requirement of a safeguard system is that an unauthorized attempt to open the canister must result in the functional destruction of the missile. Since the guidance system is the most critical -- and expensive -- component, the sealing mechanism could contain a well-aimed explosive charge, sufficient to disable the guidance section of the SLCM. A .357 magnum cartridge might suffice.

Since canisters must occasionally be opened, if only to perform scheduled maintenance, inspectors from both sides would have to unlock the PAL and note the return of the missile for repairs, modification or maintenance. When the missile was to be redeployed, the two sides would reinspect and reseal the canister.

Since the United States and the Soviet Union have already deployed many SLCMs, both conventional and nuclear, an initial inventory would be required. That inventory cannot rely upon an initial data base provided by each side, but must rest on physical inspection of ships, as well as traffic into and out of repair facilities.

Before the INF Treaty was signed, our plan would have been impossible. Now it appears to be only a small step beyond the agreed procedures for counting missiles and distinguishing the nearly identical stages of the (permitted) SS-25 from its (forbidden) sibling, the SS-20.

The United States has said it would be willing to accept limits on nuclear SLCMs if there were a way to verify those limitations. Initial explorations with both American and Soviet experts indicate that Permissive Action Links could meet that objective. It's good to know that technology can sometimes help politics manage the threats which technology itself breeds.

Peter Zimmerman is senior associate at the Carnegie Endowment for International Peace. Alton Frye is vice president and Washington director of the Council on Foreign Relations.