Governance: International coordination, security, and military use

This content reflects the field as of October 2026.

News reports sometimes announce that the world's governments have "agreed on AI." A look at three moments shows how much, and how little, that has meant.

In November 2023 the United Kingdom hosted a summit at Bletchley Park. Twenty-eight countries and the European Union put their names to a declaration. The United States and China were both among them, which was unusual for two rivals on a question of advanced technology. The declaration said of the most capable AI models: "There is potential for serious, even catastrophic, harm, either deliberate or unintentional" (Department for Science, Innovation and Technology 2023). It also said that many risks from AI are "best addressed through international cooperation."

In February 2025 France hosted the next major summit, in Paris. Its closing statement was about inclusive and sustainable AI. Dozens of countries signed, including China and India. The United States and the United Kingdom, host of the first summit, don't appear on the list of signers (Présidence de la République 2025).

In February 2026 an international panel of experts published its second report on the safety of advanced AI. The first edition, a year earlier, had carried the name of the US Department of Commerce among its backers. This time thirty countries and international organizations backed the report, among them the United Kingdom, China, and the European Union. The United States withheld its support, as TIME magazine reported (Booth 2026). The department declined to comment on why.

Set side by side, the three moments show a pattern. In 2023 the two leading AI powers could agree on a sentence about risk. By 2026 one of them wouldn't endorse an expert summary of the evidence.

The pattern has a second feature. None of the three documents required anyone to do anything. The Bletchley text uses phrases such as "we resolve" and "we affirm," and it sets no duty and no penalty. The Paris statement lists priorities. The expert report describes evidence and makes no rules. Governments were agreeing, or declining to agree, on words.

That points to a general observation about international agreement on AI. Agreement on principle has been easier to reach than agreement on anything binding, and since 2023 even agreement on principle has become harder.

Several explanations are offered, and they aren't exclusive. Governments changed, and new governments brought different priorities. Countries came to see AI as a source of economic and military advantage, which made limits look costly. Many countries outside the small group that builds advanced AI argued that the early focus on catastrophic risk ignored their concerns about access and development.

When you read that countries have reached an agreement on AI, one question sorts the report quickly: what does the agreement require of anyone? Look for who signed, what they promised to do, whether anyone can check, and what follows if a signer doesn't comply. An agreement that requires nothing can still matter as a signal of what governments are willing to say in public. It does a different job from a rule.

References

• Booth, Harry. 2026. "U.S. Withholds Support from Major International AI Safety Report." TIME, February 3, 2026.
  • Free: TIME
• Department for Science, Innovation and Technology. 2023. "The Bletchley Declaration by Countries Attending the AI Safety Summit, 1–2 November 2023."
  • Free: GOV.UK
• Présidence de la République. 2025. "Statement on Inclusive and Sustainable Artificial Intelligence for People and the Planet." Paris, February 11, 2025.
  • Free: Élysée

This content reflects the field as of October 2026.

Introduction

Since 2023 governments have met roughly once a year to discuss AI, and each meeting has ended with a document. Headlines treat these documents as agreements. They are agreements of a particular and limited kind.

This reading explains what a summit declaration is, what four summits produced, and how their emphasis moved from safety toward opportunity and impact.

What a Declaration Is

A summit is a meeting of heads of government or senior ministers from many countries. A declaration is a joint statement in which governments say what they believe or intend. A signatory is a country or organization that puts its name to such a document.

A declaration is non-binding: it creates no legal duty, and nothing happens to a signatory that ignores it. Its value lies elsewhere. It records what governments were willing to say together, it gives officials and campaigners a text to cite, and it can prepare the ground for something firmer.

Four Summits

(Department for Science, Innovation and Technology 2023; Department for Science, Innovation and Technology 2024; Présidence de la République 2025; All India Radio 2026.)

The Bletchley declaration recognized a "potential for serious, even catastrophic, harm" from the most capable models and called for international cooperation (Department for Science, Innovation and Technology 2023). The Seoul commitments moved from governments to companies. The organizations that signed promised to assess risks, set thresholds, and publish safety frameworks (Department for Science, Innovation and Technology 2024).

The Paris statement listed six priorities, including making AI accessible, making it sustainable, and ensuring it is "open, inclusive, transparent, ethical, safe, secure and trustworthy." Safety appears there as one item among several (Présidence de la République 2025).

The New Delhi declaration was organized around three themes, people, planet, and progress, and seven areas of work. One of the seven concerns the trustworthiness of AI systems. All India Radio, the Indian public broadcaster, reported that the frameworks adopted alongside it were described as voluntary and non-binding (All India Radio 2026).

The Shift in Emphasis

The names of the meetings tell the story. The first was the AI Safety Summit. The third was the AI Action Summit. The fourth was the AI Impact Summit.

Two camps read this change differently.

• It's a loss. People who work on the risks of advanced AI say the series was created to deal with severe dangers, and that the later meetings gave those dangers less room while the systems grew more capable. They point to a related sign: in 2026 the United States withheld its support from the international expert report on AI safety that the first summit set in motion (Booth 2026).
• It's a correction. Others say the first summit reflected the worries of a few rich countries and a few companies. Most of the world, on this view, cares more about access to AI, its effects on jobs, and whether its benefits reach poorer countries. India's summit put those questions first.

Both camps describe the same facts. They disagree about what the meetings are for.

What Isn't Settled

Which countries endorsed the New Delhi declaration couldn't be established for this reading. All India Radio gives a total of 88 countries and international organizations and doesn't list them. Other news reports named some major powers as endorsers, and reports differ. This reading therefore doesn't state the position of the United States, the United Kingdom, or China on that declaration.

Conclusion

As of October 2026 four summits have produced one declaration on risk, one set of company commitments, one statement on inclusive and sustainable AI, and one declaration on impact. All are non-binding. The emphasis has moved from safety toward opportunity, and the United States and the United Kingdom didn't sign the Paris statement. Whether that movement is a retreat or a broadening is disputed.

Key Terms

• Summit: A meeting of heads of government or senior ministers from many countries.
• Declaration: A joint statement in which governments say what they believe or intend.
• Signatory: A country or organization that puts its name to such a document.
• Non-binding: Creating no legal duty, so that nothing happens to a signatory that ignores it.

References

• All India Radio. 2026. "India AI Impact Summit 2026 Adopts New Delhi Declaration." February 21, 2026.
  • Free: All India Radio News
• Booth, Harry. 2026. "U.S. Withholds Support from Major International AI Safety Report." TIME, February 3, 2026.
  • Free: TIME
• Department for Science, Innovation and Technology. 2023. "The Bletchley Declaration by Countries Attending the AI Safety Summit, 1–2 November 2023."
  • Free: GOV.UK
• Department for Science, Innovation and Technology. 2024. "Frontier AI Safety Commitments, AI Seoul Summit 2024." May 21, 2024.
  • Free: GOV.UK
• Présidence de la République. 2025. "Statement on Inclusive and Sustainable Artificial Intelligence for People and the Planet." Paris, February 11, 2025.
  • Free: Élysée

This content reflects the field as of October 2026.

Introduction

Climate policy has a familiar structure: a scientific panel reports what is known, and governments negotiate on that basis. Several governments have tried to build something similar for AI, and one group of countries has gone further and written a treaty.

This reading describes two bodies that assess the science, one treaty, and the limits of each.

Assessment First

Before governments can agree on what to do about a risk, they need a shared account of what the risk is. On climate, that account comes from the Intergovernmental Panel on Climate Change, which reviews the research and states what the evidence shows.

A consensus report is a report that states what a large group of experts agree the evidence shows. Its use is to narrow factual disagreement, so that the remaining argument is about values and interests.

The International AI Safety Report

The first such effort for AI grew out of the 2023 summit at Bletchley Park. The International AI Safety Report is written by independent experts and chaired by Yoshua Bengio, a computer scientist at the University of Montreal.

Its second edition was published on February 3, 2026. TIME magazine reported that thirty countries and international organizations backed it, including the United Kingdom, China, and the European Union. The United States did not, although the US Department of Commerce had been named on the 2025 edition. Bengio told the magazine that the United States had given feedback on earlier drafts and declined to sign the final version (Booth 2026).

Many of the leading AI developers are American companies, so a shared account of the evidence is weaker without the US government's endorsement.

The United Nations Panel

A scientific panel is a standing group of experts appointed to assess the research on a subject and report on it regularly. In August 2025 the UN General Assembly created the Independent International Scientific Panel on AI by a resolution adopted without a vote.

The United Nations describes it as "the first global scientific body" on AI. It has 40 members, chosen from more than 2,600 applicants. It produces an annual report with evidence-based assessments of AI's opportunities, risks, and impacts, which is presented at a Global Dialogue on AI Governance open to all member states (United Nations n.d.).

The UN is explicit about what the panel isn't. It is "not a regulatory body" and "will not set rules, enforce standards or prescribe policy" (United Nations n.d.).

The Council of Europe Convention

The Council of Europe is an organization of 46 European countries that works on human rights. It is separate from the European Union. In 2024 it opened for signature a framework convention: a treaty that sets general principles and duties and leaves the details to each country's own laws.

The Council calls it "the first-ever international legally binding treaty in this field." Its aim is to ensure that activities across the lifecycle of AI systems are "fully consistent with human rights, democracy and the rule of law." It covers AI used by public authorities and by private actors, and it sets up a conference of the parties to follow how it's carried out. Matters of national security and national defense fall outside it (Council of Europe 2024).

Signers listed by the Council include the United States, the United Kingdom, the European Union, Canada, Japan, and Israel (Council of Europe 2024).

Signing isn't the last step. Ratification is a country's formal act of accepting a treaty as binding on itself, usually after its legislature approves. A treaty takes effect only when enough signers have ratified. The European Union ratified in May 2026, and a Council of Europe committee report dated June 2026 still spoke of working toward the convention's entry into force (Council of Europe 2026). No source checked for this reading shows the convention in force.

The Limits

Assessments inform, and they don't bind. A treaty binds only those who choose to join, and a country can sign one and never ratify it. China isn't among the convention's signers.

Conclusion

As of October 2026 two bodies assess the science of AI for governments: an expert report backed in 2026 by thirty countries and organizations without the United States, and a UN panel of 40 members created in 2025. One binding treaty exists, signed by countries that include the United States and the United Kingdom, and it wasn't shown to be in force in the sources checked. The assessment bodies supply shared facts, and none of the three compels a government that doesn't wish to be bound.

Key Terms

• Consensus report: A report that states what a large group of experts agree the evidence shows.
• Scientific panel: A standing group of experts appointed to assess the research on a subject and report on it regularly.
• Framework convention: A treaty that sets general principles and duties and leaves the details to each country's own laws.
• Ratification: A country's formal act of accepting a treaty as binding on itself, usually after its legislature approves.

References

• Booth, Harry. 2026. "U.S. Withholds Support from Major International AI Safety Report." TIME, February 3, 2026.
  • Free: TIME
• Council of Europe. 2024. Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law. CETS No. 225. Opened for signature September 5, 2024.
  • Free: Council of Europe
• Council of Europe. 2026. Steering Committee for New and Emerging Digital Technologies (CDNET), 1st Meeting, Strasbourg, 15–17 April 2026: Meeting Report. CDNET(2026)R1, June 8, 2026.
  • Free: Council of Europe (PDF)
• United Nations. n.d. "Independent International Scientific Panel on AI: Frequently Asked Questions." Accessed October 3, 2026.
  • Free: United Nations

This content reflects the field as of October 2026.

Introduction

Politicians in several countries say their nation must "win the AI race." If AI is a race, a proposal to slow down sounds like a proposal to lose. That framing is the largest single obstacle to international agreement on AI.

This reading explains the logic of the race, the main tool governments use in it, the arguments over whether rivals can coordinate anyway, and how a 2026 proposal to do so was received. It describes positions and takes none.

The Logic of a Race

Security competition is rivalry between countries over capabilities that each sees as affecting its safety and military strength. Governments in the United States and China both describe AI in these terms.

An arms race is a competition in which rivals each build up a capability because they fear the other's doing so. Its logic is simple. If one side holds back and the other doesn't, the side that held back is worse off. So each side builds, even when both would prefer that neither did.

Applied to AI: if each of two governments believes that advanced AI will bring a decisive military or economic edge, neither will accept a limit unless it's confident the other is keeping the same limit.

Export Controls as the Main Instrument

An export control is a government rule that restricts the sale of certain goods or technology to other countries. The United States has restricted sales of the most advanced AI chips to China for several years, on the reasoning that chips are the input a rival can't easily make for itself.

The policy has moved more than once. In January 2026 the US Commerce Department announced that it would review applications to export certain chips to China "on a case-by-case basis," naming one chip from Nvidia and one from AMD, if security conditions were met (Bureau of Industry and Security 2026).

China's government opposes such controls. Its 2025 action plan on global AI governance calls for cooperation and for reducing barriers to technology (Ministry of Foreign Affairs of the People's Republic of China 2025).

Can Rivals Coordinate?

The argument that they can't. AI development is hard to observe. A missile can be counted from a satellite. A training run happens inside a data center, and a model is a file that can be copied. A government can't easily confirm that a rival has stopped. A rival that cheated could gain a lasting advantage, so no government will take the risk.

The argument that they can. Rivals have limited dangerous technologies before. During the Cold War the United States and the Soviet Union negotiated limits on nuclear weapons while remaining adversaries. Those agreements were imperfect and depended on each side being able to check the other. Advocates of coordination add that advanced AI needs large numbers of specialized chips and very large data centers, which are harder to hide than software.

The historical record supports the claim that rivals can coordinate. It doesn't show that the methods carry over to AI, and that part of the argument is a forecast.

Both arguments turn on verification: checking, by someone other than the party that made a promise, that the promise is being kept.

A 2026 Proposal and Its Reception

In September 2026 Dario Amodei, chief executive of the AI company Anthropic, published an essay arguing that companies and governments should slow the rate at which AI capabilities improve (Amodei 2026b). He leads a company that would be bound by such an arrangement and that would also benefit if rivals were bound.

The essay treats the international problem directly. It argues that democratic countries should coordinate among themselves first and keep their lead over China, including through limits on chip sales. On agreements with China, it describes several levels, from narrow bans to a full slowdown, and compares one of them to Cold War arms treaties. It warns that any agreement "must be limited enough that defection would not be militarily existential" unless it can be verified with near certainty. It judges a full slowdown with China "unlikely to actually happen any time soon" (Amodei 2026b).

The reception was divided. The technology news site The Register described the plan as a truce that would let the largest companies compete less fiercely, and reported political resistance: the Speaker of the US House of Representatives warned that it could "smother innovation," and President Trump said the United States must lead (The Register 2026).

Conclusion

Governments that treat AI as a source of military and economic advantage are reluctant to accept limits that a rival might not keep. As of October 2026 the main instrument of the competition is control over advanced chips, and US policy on those controls has changed repeatedly. Whether rivals can coordinate on AI as they once did on nuclear weapons depends on verification, and nobody has yet shown a method that both sides would trust.

Key Terms

• Security competition: Rivalry between countries over capabilities that each sees as affecting its safety and military strength.
• Arms race: A competition in which rivals each build up a capability because they fear the other's doing so.
• Export control: A government rule that restricts the sale of certain goods or technology to other countries.
• Verification: Checking, by someone other than the party that made a promise, that the promise is being kept.

References

• Amodei, Dario. 2026b. "We Must Pace the Frontier." September 2026.
  • Free: Dario Amodei's website
• Bureau of Industry and Security. 2026. "Department of Commerce Revises License Review Policy for Semiconductors Exported to China." Press release, January 13, 2026.
  • Free: Bureau of Industry and Security
• Ministry of Foreign Affairs of the People's Republic of China. 2025. "Global AI Governance Action Plan." July 26, 2025.
  • Free: Ministry of Foreign Affairs of the People's Republic of China
• The Register. 2026. "Big AI Sets Out Its Terms for Regulatory Capture and Calls It 'Pace the Frontier'." September 14, 2026.
  • Free: The Register

This content reflects the field as of October 2026.

Introduction

"Killer robots" is the phrase headlines use. The dispute behind it concerns how much of the decision to use force may be handed to a machine, and whether new international law is needed to settle that.

This reading sets out a military's own policy, a humanitarian organization's proposal, the votes of governments at the United Nations, and a 2026 dispute between a government and an AI supplier. It takes no position on what the rules should be.

What Is Being Argued About

An autonomous weapon system is a weapon that, once activated, selects and attacks targets without further human intervention. Advances in AI widen the range of targets and settings in which a machine could be left to choose.

States disagree about whether such weapons should be banned, restricted by new rules, or governed by the laws of war that already exist.

The US Directive

The US Department of Defense sets its policy in a directive updated in January 2023. The central sentence reads: "Autonomous and semi-autonomous weapon systems will be designed to allow commanders and operators to exercise appropriate levels of human judgment over the use of force" (US Department of Defense 2023, sec. 1.2).

The standard is human judgment: the requirement that people decide, with care, when and how force is used, even if a machine carries out the attack. The directive permits autonomous weapons and requires senior officials to approve them before development and again before fielding, with some categories exempt. The word "appropriate" leaves the level of human involvement to be decided case by case.

The Red Cross Position

The International Committee of the Red Cross, a humanitarian organization with a recognized role under the laws of war, recommended in 2021 that states adopt new rules in three parts (International Committee of the Red Cross 2021).

• "Unpredictable autonomous weapon systems should be expressly ruled out."
• "Use of autonomous weapon systems to target human beings should be ruled out."
• All others should be regulated, with limits on targets, duration, area, and scale of use, and with human supervision.

In October 2023 its president, Mirjana Spoljaric, and the UN Secretary-General, António Guterres, asked world leaders to negotiate "a new legally binding instrument" and "to conclude such negotiations by 2026" (Guterres and Spoljaric 2023). A binding instrument is a treaty or similar agreement that creates legal duties for the states that join it.

The UN Votes

The UN General Assembly has voted on resolutions about these weapons. Such resolutions aren't binding law.

(United Nations 2024; United Nations 2025.)

A large majority of states supports international work on the issue, and several states with advanced militaries voted against or abstained. The 2025 resolution speaks of completing the elements of an instrument "with a view to future negotiations" (United Nations 2025). No source checked for this reading shows a treaty concluded by the 2026 deadline.

A 2026 Dispute Between a Government and a Supplier

A usage restriction is a limit that a supplier writes into its contract or terms on what a customer may do with its product.

According to the defense news site Defense One, Anthropic's agreement with the US Defense Department prohibited using its AI for domestic mass surveillance and for fully autonomous weapons. The department sought to change the contract to permit "all lawful use." The company refused, and in February 2026 the defense secretary designated it a supply-chain risk, which bars it from military work (Croxton 2026).

Other suppliers took different terms. NPR reported that Google and xAI had agreed to allow use of their AI in any lawful scenario, and that OpenAI announced an agreement with the department on the day of the designation (Brumfiel 2026). TechCrunch later reported that Google's contract says the company doesn't intend its AI to be used for domestic mass surveillance or autonomous weapons, that OpenAI's has comparable language, and that it was unclear whether Google's provisions can be enforced (Bort 2026).

Litigation followed. In August 2026 a federal court in California held one of the department's two legal grounds for the designation unlawful. On September 25, 2026 a federal appeals court in Washington upheld the other by two votes to one. The company said it disagreed and was considering further review. A defense official said the ruling confirmed that the company is a risk to the defense supply chain (Croxton 2026).

The department's position is that it should be able to use a supplier's product for any lawful purpose. The company's position is that it won't supply its AI for those two uses. The dispute wasn't finished at the time of writing.

Conclusion

As of October 2026 no binding international ban or restriction on autonomous weapons is shown in the sources checked. The US directive requires appropriate levels of human judgment and permits such weapons. The Red Cross and the UN Secretary-General have asked for a treaty that would prohibit some and restrict the rest. Most states have voted for further UN work, and several major military powers have voted against or abstained.

Key Terms

• Autonomous weapon system: A weapon that, once activated, selects and attacks targets without further human intervention.
• Human judgment: The requirement that people decide, with care, when and how force is used, even if a machine carries out the attack.
• Binding instrument: A treaty or similar agreement that creates legal duties for the states that join it.
• Usage restriction: A limit that a supplier writes into its contract or terms on what a customer may do with its product.

References

• Bort, Julie. 2026. "Google Expands Pentagon's Access to Its AI After Anthropic's Refusal." TechCrunch, April 28, 2026.
  • Free: TechCrunch
• Brumfiel, Geoff. 2026. "OpenAI Announces Pentagon Deal After Trump Bans Anthropic." NPR, February 27, 2026.
  • Free: NPR
• Croxton, John. 2026. "Anthropic Loses Legal Fight to Shed DOD's Designation as a 'Supply-Chain Risk'." Defense One, September 25, 2026.
  • Free: Defense One
• Guterres, António, and Mirjana Spoljaric. 2023. "Joint Call by the United Nations Secretary-General and the President of the International Committee of the Red Cross for States to Establish New Prohibitions and Restrictions on Autonomous Weapon Systems." October 5, 2023.
  • Free: International Committee of the Red Cross
• International Committee of the Red Cross. 2021. "ICRC Position on Autonomous Weapon Systems." May 12, 2021.
  • Free: International Committee of the Red Cross
• United Nations. 2024. "General Assembly Adopts Resolution on Lethal Autonomous Weapons Systems." Press release GA/12660, December 2, 2024.
  • Free: United Nations
• United Nations. 2025. Press release GA/12736 on the General Assembly's adoption of First Committee texts, including on lethal autonomous weapons systems.
  • Free: United Nations
• US Department of Defense. 2023. DoD Directive 3000.09: Autonomy in Weapon Systems. January 25, 2023.
  • Free: US Department of Defense (PDF)

Introduction

In October 2023 the head of the United Nations and the head of the Red Cross jointly asked governments to write new international law on autonomous weapons by 2026. The text is short, and its wording differs in instructive ways from the policy of the US military.

This reading goes through the call, sets it beside the US Defense Department's directive, and asks what has happened as the deadline arrived.

Locating the Passage

The text is on the website of the International Committee of the Red Cross, free and listed in the References. Its full title is "Joint call by the United Nations Secretary-General and the President of the International Committee of the Red Cross for States to establish new prohibitions and restrictions on Autonomous Weapon Systems." It is dated October 5, 2023 and runs to about eleven short paragraphs. The paragraphs aren't numbered, so this reading identifies passages by quoting them. All quotations in the first three steps are from this text (Guterres and Spoljaric 2023).

The authors are António Guterres, Secretary-General of the United Nations, and Mirjana Spoljaric, president of the International Committee of the Red Cross, a humanitarian organization with a recognized role in the laws of war. Neither can make international law. The call is addressed to states because only states can negotiate a treaty.

Walking Through the Passage

Step 1: Read what the two leaders ask states to do, and by when

The opening sentence says the two are "calling on political leaders to urgently establish new international rules on autonomous weapon systems, to protect humanity."

The specific request comes later: "We call on world leaders to launch negotiations of a new legally binding instrument to set clear prohibitions and restrictions on autonomous weapon systems and to conclude such negotiations by 2026."

Three things are being asked: to start negotiating, to make the result "legally binding," which means a treaty and not a declaration, and to finish by 2026. That gave states a little over three years.

The call also says what it means by the term. It describes these as "weapon systems that select targets and apply force without human intervention."

Step 2: Identify the two kinds of rule requested

The title names both: "prohibitions and restrictions." They're different instruments. A prohibition bans a category outright. A restriction allows a category under conditions.

On prohibitions, the call says: "This means prohibiting autonomous weapon systems which function in such a way that their effects cannot be predicted." It gives an example: "allowing autonomous weapons to be controlled by machine learning algorithms – fundamentally unpredictable software which writes itself – is an unacceptably dangerous proposition."

That sentence makes a strong technical claim. Machine learning systems are built by training on examples, and their behavior in new situations can't be fully predicted from their design. The call treats that property as disqualifying for a weapon. Those who favor such systems would answer that testing can establish how reliably a system behaves within set conditions.

The call also draws a line around targeting people: "The autonomous targeting of humans by machines is a moral line that we must not cross."

On restrictions, it says that "clear restrictions are needed for all other types of autonomous weapons." These include "limiting where, when and for how long they are used, the types of targets they strike and the scale of force used, as well as ensuring the ability for effective human supervision, and timely intervention and deactivation."

Step 3: Read the reasons given

The call says that autonomous weapon systems "pose serious humanitarian, legal, ethical and security concerns." Each adjective points to a separate argument.

• Humanitarian. A weapon that chooses its own targets may strike civilians.
• Legal. Those laws assume a human who can be held responsible. If a machine selects the target, it's unclear who answers for an unlawful strike.
• Ethical. The call states: "Human control must be retained in life and death decisions."
• Security. The call warns that by "creating a perception of reduced risk to military forces and to civilians, they may lower the threshold for engaging in conflicts, inadvertently escalating violence."

The ethical reason is of a different kind. The other three are claims about consequences, which better technology could in principle answer. The ethical reason says that some decisions shouldn't be delegated however well the machine performs.

Step 4: Set the call beside the US directive and mark the difference

The US Department of Defense states its policy this way: "Autonomous and semi-autonomous weapon systems will be designed to allow commanders and operators to exercise appropriate levels of human judgment over the use of force" (US Department of Defense 2023, sec. 1.2).

Compare the key phrases. The call says "human control must be retained." The directive says "appropriate levels of human judgment."

Control and judgment aren't the same. Judgment can be exercised before a weapon is launched: a commander decides that conditions are right, sets the area and the kind of target, and activates the system. After that the machine may act with nobody able to stop it, and the standard is still met. Control suggests a person who can supervise and intervene while the system operates, which is what the call's "timely intervention and deactivation" asks for.

"Appropriate" matters too. It makes the required level depend on the situation, as judged by the military itself.

The two also differ in kind. The directive is one country's internal policy, which that country can change. The call asks for limits fixed in law that binds every state that joins.

Step 5: Note that the deadline has arrived and say what the UN votes show

The call's deadline was 2026. The UN General Assembly's votes show how far states have moved.

In December 2024 the General Assembly adopted a resolution on these weapons by 166 votes to 3, with 15 abstentions (United Nations 2024). In 2025 it adopted another by 164 votes to 6, with 7 abstentions. The United States was among the six voting against. That resolution speaks of completing the elements of an instrument "with a view to future negotiations" (United Nations 2025).

So a large majority of states supports continued work. The phrase "future negotiations" indicates that formal treaty negotiations hadn't begun in 2025. As of October 2026 no source checked for this reading shows a treaty. Several states with advanced militaries voted against or abstained, and a treaty they don't join wouldn't bind them.

Key Considerations

A call is an appeal. It isn't an agreement, and no state signed it. Reports that "the UN called for a ban" are accurate about the Secretary-General and say nothing about what member states have decided.

A common mistake is to treat "human judgment" and "human control" as the same standard. News reports often use the phrases interchangeably. The difference is the substance of the dispute. A government can honestly say it requires human judgment over the use of force while fielding weapons that select their own targets after launch.

The call also asks for less than a total ban. It would prohibit two categories and restrict the rest.

Summary

The joint call asks states for a binding treaty that prohibits some autonomous weapons and restricts the others, and the US directive sets an internal standard that permits them under review. The result of the reading is this comparison.

References

• Guterres, António, and Mirjana Spoljaric. 2023. "Joint Call by the United Nations Secretary-General and the President of the International Committee of the Red Cross for States to Establish New Prohibitions and Restrictions on Autonomous Weapon Systems." October 5, 2023.
  • Free: International Committee of the Red Cross
• United Nations. 2024. "General Assembly Adopts Resolution on Lethal Autonomous Weapons Systems." Press release GA/12660, December 2, 2024.
  • Free: United Nations
• United Nations. 2025. Press release GA/12736 on the General Assembly's adoption of First Committee texts, including on lethal autonomous weapons systems.
  • Free: United Nations
• US Department of Defense. 2023. DoD Directive 3000.09: Autonomy in Weapon Systems. January 25, 2023.
  • Free: US Department of Defense (PDF)