The Treaty That Bans
Biological Weapons
but Cannot Enforce It.

The Convention on the Prohibition of the Development, Production and Stockpiling of Bacteriological (Biological) and Toxin Weapons and on Their Destruction, universally abbreviated to the Biological Weapons Convention or BWC, was opened for signature on April 10, 1972. It entered into force on March 26, 1975. It was the first multilateral treaty to ban an entire category of weapons of mass destruction, and it did so in sweeping terms: Article I prohibits development, production, stockpiling, or acquisition of biological agents for offensive purposes under any circumstances.

The BWC currently has 183 states parties. Every major power is a signatory.

The Soviet Union operated Biopreparat, a massive clandestine biological weapons program, throughout its period as a BWC party. At its height, Biopreparat employed more than 60,000 scientists in dozens of facilities developing weaponized anthrax, smallpox, plague, viral hemorrhagic fevers, and other agents. It continued into the 1990s, after the treaty had been in force for nearly two decades. The program was concealed from BWC review meetings, from diplomatic contacts, and from intelligence agencies for years. It was ultimately revealed not by any treaty mechanism but by defectors, most prominently Dr. Kanatjan Alibekov, who later published an account of his time as a senior Biopreparat official under the name Ken Alibek.

The lesson for AI governance is precise and uncomfortable: a treaty that prohibits an activity without creating a system for detecting violations is not a governance mechanism. It is a restatement of shared norms without any means to enforce them.

What the BWC does and does not do

The treaty's prohibitions are clear and broad. It bans development, production, and stockpiling of biological weapons. It requires destruction of existing stockpiles. It prohibits transfer of biological weapons capability to other states. These obligations are legally binding on all state parties.

The consultation mechanism in Article V allows a state party to request consultation with other parties if it believes a violation has occurred. In practice, this means one country can formally ask another country whether it is violating the treaty, and the accused country can decline to answer in any meaningful way. The mechanism has been invoked rarely and has never produced a finding of violation or any remedy.

There is no standing verification body. There are no inspectors. There is no requirement to declare biological research programs beyond voluntary confidence-building measures that most states do not fully complete. The treaty has five-year review conferences where states parties discuss implementation, but these are diplomatic meetings, not investigative processes.

Why verification was never included

The BWC was negotiated quickly relative to other arms control treaties of the era — the negotiating text was developed between 1969 and 1972. The speed reflected the political priority of establishing the prohibition principle in international law, at a moment when both the US and Soviet Union were willing to accept one. President Nixon unilaterally renounced US biological weapons in 1969 and ordered US stockpiles destroyed before the treaty was concluded.

The verification question was regarded as too technically difficult and politically contentious to resolve in the same negotiating window. Biological agents are produced using the same equipment used in pharmaceutical and vaccine manufacturing. The dual-use problem was even more acute for biological weapons than for chemical weapons, because the relevant industrial base is inseparable from civilian medicine in a way that specialized chemical weapons precursors are not. Designing inspection protocols that could verify the absence of a weapons program without revealing sensitive pharmaceutical or biodefense research was not resolved during the negotiating period, and the decision was made to proceed with the prohibition without the verification.

That decision was understandable given the political conditions. It was also, in retrospect, a serious error.

The Soviet program and what it proved

Biopreparat operated continuously from before the BWC's entry into force through the Soviet collapse and into the early years of the Russian Federation. Alibekov's account, published in 1999 as Biohazard, describes a program structured explicitly to evade any potential inspection: biological agents were classified under civilian cover names, facilities were presented as pharmaceutical plants, and senior officials maintained careful separation between the treaty-compliant civilian face of Soviet biological research and the weapons programs that ran in parallel.

This was not a marginal violation. It was a systematic, large-scale, organizationally sophisticated program that operated in direct contradiction of the Soviet Union's treaty obligations for approximately fifteen years. The Soviet government knew it was violating the treaty. Soviet officials participated in BWC review conferences and confidence-building measure submissions during this period. The treaty provided no mechanism to detect, investigate, or respond to any of this.

The United States was aware through intelligence that Soviet biological weapons activity continued after the treaty's entry into force, but the absence of a formal inspection mechanism made formal attribution and response impossible within the treaty framework. Bilateral diplomatic confrontations occurred; they produced Russian acknowledgment of a "past" program and denials of current activity, with no independent verification possible.

The failed attempt to add verification

From 1994 to 2001, a multilateral working group negotiated a draft verification protocol intended to add inspection rights to the BWC. The Ad Hoc Group of states parties met twenty-four times over seven years and produced a draft protocol of more than two hundred pages. In July 2001, the United States rejected the draft. The principal US objections were that the protocol's inspection regime would not effectively detect sophisticated violations while imposing unacceptable burdens on US pharmaceutical companies and exposing classified US biodefense programs to inspection by states that might be BWC violators themselves.

Both objections reflected real concerns about an imperfect verification design. Neither addressed the underlying problem, which was that the treaty continued to operate without any verification mechanism. After the US rejection, negotiations effectively collapsed, and no subsequent negotiating mandate has been agreed. The BWC entered its sixth decade as a comprehensive prohibition with no enforcement architecture.

The direct lesson for AI governance

The BWC failure is the most important negative case study for AI governance design. It demonstrates three things with high confidence.

First, a prohibition without verification is not governance. The breadth of the prohibition in the BWC — which bans an entire weapons category with near-universal participation — did not prevent the Soviet Union from operating one of the largest biological weapons programs in history. The words of the treaty created no obligation that the Soviet government chose to observe when it calculated that non-compliance was in its interest and undetectable. Words in treaty text, however clearly written and universally ratified, do not create behavioral constraints in the absence of monitoring and consequences for detection.

Second, verification cannot be added after the fact when major powers have concluded it is not in their interest. The 1994-2001 negotiation failed not because a verification architecture was technically impossible, but because the major power most capable of designing effective verification concluded that the available design served its competitors' interests more than its own. Once a treaty exists without verification, the political conditions for adding verification may never return. For AI governance, this means that the verification architecture must be agreed as part of the original framework, not deferred as a technical detail to be resolved later.

Third, the dual-use problem is a design challenge, not a treaty-blocking obstacle. The BWC's verification failure was substantially attributed to the dual-use nature of biological technology. AI development faces an analogous dual-use challenge, arguably more acute. But the Chemical Weapons Convention successfully designed verification for a dual-use technology by carefully scoping what inspectors could and could not access. The lesson from the CWC's success is not that dual-use problems prevent verification. It is that solving them requires sustained technical and diplomatic work that the BWC negotiators did not undertake before the prohibition was concluded.

"An AI governance treaty without verification mechanisms would be the BWC of the 21st century: broad enough to create the impression of governance, weak enough to constrain no one who chooses non-compliance."

Naoto Nakada, Founder · Nakada Foundation to Save Humanity

The Biological Weapons Convention is not a failure of intent. The prohibition norms it established are genuine and have constrained many states from developing biological weapons programs they might otherwise have pursued. It is a failure of design: a treaty that established the wrong it sought to prevent in clear terms, and then created no mechanism to detect that wrong when states chose to commit it. AI governance proposals that establish prohibitions on dangerous AI development without specifying how those prohibitions would be verified are following the same design path. The BWC's history is a warning, not a model.