To an extent, such a reaction is understandable. Western societies are frustrated with the slow pace of AI governance in their own countries. In the US, there seems to be so much focus on getting ahead in the AI race that the administration frequently portrays AI legislative proposals as measures that could hamper the country’s national interests3. In the EU, while there is already the first comprehensive AI law, governance is slow because of the emphasis on fundamental rights, the need to secure approval from a very large group of stakeholders, take into account US commercial interests, and respond to a new narrative suggesting the need to innovate or become irrelevant. In fact, in the EU, the process is often so slow that by the time a law is finally adopted, the technology has already advanced to the point where new rules are needed4.

Against this background, China becoming the first country to enact a dedicated law specifically targeting anthropomorphic AI was perceived as a progressive step forward in AI policy.

The mechanisms worthy of praise

To be fair, the law does contain legal mechanisms worthy of praise. For example, it contains mandatory design rules instead of vague principles. Among other things, the law requires providers to clearly indicate to users that they are interacting with artificial intelligence rather than a natural person. It also requires providers, when a user shows signs of overdependence or logs in for the first time, to remind them (through pop-up windows or other means) that the interactive content is generated by artificial intelligence. The law also specifically acknowledges that minors and the elderly are among the groups most vulnerable to AI anthropomorphism and establishes concrete, vulnerability-specific measures to protect them5. The media has even reported that, following the introduction of the new rules, ByteDance and Alibaba (some of the country’s most widely used AI applications) had to disable human-like AI custom agents6.

Subscribe now

Protecting the regime first, users second

However, while reading the law, I often found myself asking what legislators had in mind when drafting it. I have not seen any documents accompanying the law, but after familiarizing myself with it, I was left with the impression that it is as much about reinforcing control over society as it is about protecting vulnerable people from AI harms. The structure of the law itself contributes to this impression. Article 8 refers to seven different categories of content that shall not be generated. And no, the law does not begin with protecting minors or other vulnerable groups. It begins with protecting China’s political system. The prohibited content includes, among other things, material concerning the subversion of state power and the overthrow of the socialist system, violations of the core socialist values, illegal religious activities, and more.

Article 8: Providers of anthropomorphic interactive services shall comply with laws and administrative regulations, respect social morality and ethics, and shall not engage in the following activities: (i) Generating content that endangers national security, honor and interests, incites the subversion of state power and the overthrow of the socialist system, incites the division of the country and undermines national unity, promotes terrorism, extremism and historical nihilism, violates the core socialist values, carries out illegal religious activities, promotes ethnic hatred and ethnic discrimination, incites group antagonism, disseminates obscenity, pornography, gambling, violence or incites crime, spreads rumors, insults or defames others, or infringes on the legitimate rights and interests of others7.

Furthermore, the law establishes a mechanism that could, should there be political will, allow users to be tracked down. Article 12 requires providers of anthropomorphic interactive services to sign service agreements with users, require them to register in accordance with the law and the agreement, and provide necessary information such as the user’s age, guardian, or emergency contact person8. It is not difficult to imagine that the Chinese Communist Party would have little difficulty accessing such information.

The global picture

I would love to believe that I am wrong and that the Chinese leadership is simply in a rush to protect its people from AI harms. However, both my experience of growing up in Eastern Europe, where autocratic methods of governance were not unheard of, and China’s track record on human rights9 suggest otherwise.

Indeed, while China has an interest in keeping its population healthy and competitive, its regime may have no less interest in its own survival. AI promises not only to harm ordinary people but also, potentially, the ruling elites. Whether intended or not, the law provides another legal instrument that can be used to moderate content, reinforce the government’s grip over society, and punish dissent.

In the same week that China’s law on anthropomorphic AI came into effect, Xi Jinping launched a new AI alliance. The World Artificial Intelligence Cooperation Organisation (WAICO) is a new coalition of 29 nations that plans to develop AI regulation across countries10.

As China seeks to shape AI policy globally, we should look at its own laws not only with enthusiasm for how quickly they address regulatory gaps, but also with an understanding that those gaps may be filled in ways that reflect and reinforce an authoritarian model of governance.

Author’s Note on AI Collaboration

The content and ideas in this article are entirely the work of the author. Anthropic’s Claude, Chat GPT and Google Gemini were used solely as light editorial tools to verify grammar, organize the footnotes and select a suitable graphic for the text. All creative direction, and factual verifications were performed exclusively by the human author.

Leave a comment

2

Luiza’s Newsletter. (2026). The world’s strictest law on human-like AI just took effect. Here is what it means.

Luiza's Newsletter

The World's Strictest Law on Human-Like AI

Read more

a month ago · 37 likes · 5 comments · Luiza Jarovsky, PhD

Against this background, I started thinking about Ukraine, and in particular, about two of the most defining challenges the country faces in its war against Russia: when and under what conditions to end the war, and who should fight it. These are questions where public opinion is deeply fractured, and where I can clearly see the limits of digital democracy.

Who gets to vote on peace?

As far as e-democracy is concerned, Ukraine’s performance is impressive2. The state-run Diia app allows citizens to vote on issues of national significance. For example, as a result of a vote on the Diia app, Ukrainians changed the date when Christmas is celebrated from January 7th to December 25th3. It should be noted, though, that the platform’s polling and surveys are not yet AI-powered.

So, what if an AI-powered tool, similar to those used in Taiwan or Bowling Green, were used to help the people of Ukraine arrive at a common position on how the war should end?

Imagine Ukraine tries to put a peace deal with Russia to a vote using such a tool. On its face, this looks like a reasonable step: give the people who bear the burden of the war and die in it the ability to decide when to sign a deal, and on what conditions. But the first serious challenge is deciding who should vote.

Should it be only those Ukrainians currently living in Ukraine, who face the war and its consequences firsthand? Or all Ukrainians, including those who are abroad?

This is something many abstain from discussing because of the sensitivity of the issue. Many Ukrainians have settled in North America and Europe over almost five years of war. Some genuinely want to return to their home country, but many do not. The end of the war might mean their temporary residence permits and work authorizations expire, requiring them to go back. This does not make those who don’t want to return unpatriotic. It is simply a reality of life.

However, it does mean that Ukrainians can be driven by structurally different incentives when it comes to the question of ending the war. Those living inside Ukraine may want a swifter end, even one that involves painful concessions on sovereignty and territorial integrity. Meanwhile, those living outside, who do not feel the daily weight of the conflict, may favor a longer war aimed at the full restoration of Ukrainian territory.

If this scenario occurs, I do not think the government should extend the vote on ending the war to Ukrainians abroad. It would, in my view, be fairer to allow only those who immediately bear the burden of the war to decide how long it should last. Therefore, I fail to see an act of direct democracy (powered by AI and covering all Ukrainians both inside and outside the country) as a viable mechanism to answer how and under what conditions the war should end.

That said, this does not mean there is no scope for AI in this domain. The Carl Miller and Beth Goldberg’s article mentions Jigsaw’s Sensemaking4. This is an AI tool designed to help policymakers gather and understand public opinion at scale. In my opinion, the Ukrainian government could use something like this to better understand what society thinks about how the war should end, without requiring citizens to state their views publicly. This matters because, during wartime, many people would hesitate to reveal their true attitudes for fear of being labeled insufficiently patriotic. The Habermas Machine5, developed by Google and mentioned in the article, can also act as a mediator between people discussing divisive topics and could help identify common ground among Ukrainians (both inside and outside the country) on what a just end to the war might look like.

There is thus real scope for AI tools in understanding what Ukrainians think. I am, at the same time, more skeptical of using them to run a direct-democracy-style vote open to all Ukrainian citizens, regardless of where they live.

Subscribe now

AI and military recruitment

As Carl Miller and Beth Goldberg note, trust is never a minor issue when it comes to AI and digital democracy6. This is especially so when it comes to the question of military recruitment.

These days, Ukraine (and to a large extend Russia) faces a shortage of manpower. The government has had no choice but to call hundreds of thousands of men to the frontline. The Constitution of Ukraine7 and other laws mandate that Ukrainian men, unless they fall under certain exceptions, must protect their homeland. I will not take a moral stance on this. The real cost of a “just” end to the war may be such that there are legitimate reasons not to take very patriotic stances on the matter, and the circumstances of this conflict differ from the wars that tend to get romanticized. I do believe that Ukrainians, on the whole, are very patriotic and courageous people and that the Ukrainian government, including its Ministry of Defense and Ministry of Digital Transformation, is doing more than what might be considered possible.

The question is whether AI-powered tools should be used to help the government recruit more people into the military. This was discussed in Ukrainian policy circles. It was even proposed to allow military service subpoenas to be sent via the already mentioned Diia app. The Ministry of Digital Transformation refused8 for the reason that using the platform for this purpose (however legitimate) would undermine public trust in Diia as a whole. If citizens come to associate e-governance with conscription notices, they will trust it less for everything else.

In my opinion this case is very illustrative in that sometimes using AI for egovernance purposes might undermine trust in technologies even when the reason for so doing is legitimate in itself.

At the same time, there are cases where AI can strengthen, rather than erode, trust in military-related services. Very recently, a new service called For Defenders and Their Families was launched on the Diia portal, allowing military personnel, veterans, and families of fallen defenders to submit a single application to access multiple government services at once9. One can imagine agentic AI being deployed in the future to make this process even more efficient.

Share

Conclusion

As illustrated by Carl Miller and Beth Goldberg in their AI Policy Perspectives article AI-powered tools can be genuinely useful in helping governments understand public opinion, find common ground, and deliver services more effectively. However, looking at it from the perspective of the war in Ukraine where very sensitive decisions need to be made as to how to end the war and who should die in it, I also see the challenges that the above authors mention. Direct democracy using AI tools does not always appropriately address the situation and public trust does not always sit well with the AI tools.

Author’s Note on AI Collaboration

The content and ideas in this article are entirely the work of the author. Anthropic’s Claude and Google Gemini were used solely as light editorial tools to verify grammar and improve the overall flow of the prose. All creative direction, structural decisions, and factual verifications were performed exclusively by the human author.

Leave a comment

1

AI Policy Perspectives. (2026, June 17). Stop Shouting. Start Policymaking.

AI Policy Perspectives

Stop Shouting. Start Policymaking.

Guest post from Carl Miller, a technologist and writer based at the think tank Demos in London, and Beth Goldberg, head of R&D at Jigsaw, a Google technology incubator, who teaches at Yale…

Read more

2 months ago · 15 likes · 1 comment · AI Policy Perspectives

A week ago, elections took place in neighboring Hungary. They attracted global attention: at stake was whether Viktor Orbán, the long-serving Prime Minister whose government cultivated close ties with Putin’s Russia while clashing increasingly with Brussels, would survive the most serious challenge of his sixteen years in power. Coverage mostly focused on the fight between left and right, Eurosceptics and Euro-optimists, democratic and authoritarian governance models2.

What received far less attention was that here too, AI-driven disinformation played a significant and documented role. TikTok removed four covert influence networks in the months before the vote, most of which spread content favorable to Orbán’s Fidesz party3. Fidesz itself deployed AI-generated videos as part of its official campaign, including a fabricated clip depicting the execution of a Hungarian soldier, designed to stoke fears about Magyar’s alleged positions on Ukraine4. Unlike Romania, this didn’t dominate the global tech conversation. But it was perhaps only because Orbán, the side that deployed disinformation most aggressively, lost anyway. This gave me reason to think about a question that sounds simple but isn’t: what does it actually mean to make a democratic choice?

When I think about democracy, I imagine a system where major decisions are made by the people. But that carries a hidden assumption that people are in a position to decide well. In my legal education, access to full information is treated as a precondition for sound judgment. A decision based on manipulated information has a higher chance of being wrong, not because the person is foolish, but because the inputs were corrupted.

Should we apply the same lens to AI-driven disinformation? If the information environment during an election is systematically distorted, through fabricated content, coordinated bot networks, and AI-generated deepfakes, are voters making free choices, or are they deciding while effectively intoxicated?

Subscribe now

The Romanian Constitutional Court addressed a version of this question. Its reasoning led to the conclusion that the democratic process was compromised and that the integrity of the citizens' votes and the fairness of the competition were distorted by digital manipulation5. Many disagreed and argued that it was the Court itself, not the algorithm, that breached democratic principles, the ruling was made ex officio, relied heavily on intelligence reports rather than conventional legal evidence, and came while overseas voting was already underway6. Whatever one thinks of the outcome, the underlying question is worth taking seriously: was the manipulation pervasive enough to trigger a systemic failure of the information ecosystem?

If yes, we have to ask why technologies marketed as tools for society’s benefit keep undermining its democratic foundations. The answer is that the platforms are simply not built to care, and this is not an accident but a peculiarity of the business model. TikTok’s algorithm is optimized for user retention7. In that setup, a state-sponsored deepfake and a genuine political debate look identical: both are measured by how long they keep users on the app. The algorithm can’t tell the difference, and it has no financial incentive to try. This is the misalignment at the heart of the problem: a platform’s commercial interest in maximizing engagement runs directly against society’s interest in an accurate information environment. In a certain sense, we have handed the architecture of our democratic discourse to a system that treats a national security threat as highly engaging content.

Solutions are being tried. Platforms propose mandatory labeling of AI-generated political content and digital watermarking to verify media origins8. The European Commission activated its Digital Services Act rapid response mechanism ahead of the Hungarian vote9. These efforts do bear fruits. But as long as the commercial interests of platforms and the civic interests of citizens remain misaligned, there will always be new ways to cause democratic harm. Not out of malice, but out of indifference. What is needed is not just better tools, but a new “social contract” between BigTech and the public.

Hungary’s story shows what happens when a government loses the trust of its own people. Years of mismanagement, corruption, and broken promises gradually eroded the unspoken contract between Orbán’s regime and Hungarian society, and when that contract finally broke, so did his grip on power. Tech companies should take note. Like governments, they operate on borrowed trust. And as Hungary shows, by the time people take to the streets, or to the polls, it is already too late.

Author’s Note on AI Collaboration

The content and ideas in this article are entirely the work of the author. Anthropic’s Claude was used solely as a light editorial tool to verify grammar and improve the overall flow of the prose. All creative direction, structural decisions, and factual verifications were performed exclusively by the human author.

Leave a comment

The signing ceremony of the Black Sea Grain Initiative at Dolmabahçe Palace, Istanbul, Türkiye, 22 July 2022. Photo: UNIC Ankara/Levent Kulu

As artificial intelligence systems become more capable, proposals emerge to build international safety guarantees based on arms control precedents. Leading voices in the AI field such as OpenAI CEO Sam Altman2, UN Secretary-General António Guterres3 and Google DeepMind’s Demis Hassabis4 have called for an “IAEA for AI” and “CERN for AI”. The institution would serve as a neutral entity that could inspect systems, require audits, and ensure compliance with safety standards between rival powers like the United States and China.

The idea is that international guarantees of this kind could help slow down dangerous capability races without requiring either rival state to give up its technological ambitions. Technical proposals that have already been put forward include compute governance tracking specialized AI chips5, data center inspections, hardware-enabled location verification through network latency, and developer audits.

And yet, recent experience from a somewhat different but nonetheless relevant context suggests caution. When vital interests diverge fundamentally, power and political incentives can overwhelm even well-designed international agreements.

When Verification Worked But Power Prevailed

The Russian full-scale invasion of Ukraine endangered millions of people worldwide who depended on Ukrainian grain supplies. It was estimated that because Ukraine’s grain exports were blocked, at least 47 million people (mainly from developing countries) were facing acute food insecurity6. Given that the two countries were at war with very little communication between them (reduced to matters such as prisoner exchange) and zero trust, an arrangement was proposed whereby a third (neutral) party would serve as an intermediary, allowing agricultural goods to be exported out of Ukraine. No deal was signed between Russia and Ukraine. Instead, the two countries signed separate agreements with the UN and Türkiye providing for the safe operation of vessels and the export of food. The initiative established a Joint Coordination Centre in Istanbul with Russian, Ukrainian, Turkish, and UN representatives, along with joint inspection teams, demined maritime corridors, and registration procedures7.

Initially, the arrangement worked. Nearly 33 million metric tons of grain moved through Ukrainian ports8. Inspections functioned smoothly. Turkey and the UN stayed neutral. The technical verification regime was functioning too. Elements similar to those proposed for the IAEA-for-AI, such as neutral third parties, clear procedures, objective monitoring, existed and operated effectively.

And then, on July 17, 2023, despite many elements of the arrangement (such as inspections) working as planned, Russia unilaterally withdrew from it9. International criticism followed, but it didn’t change the Russian stance. What’s more, no penalties were imposed on Russia. The failure was political, not technical. Russia simply stopped believing it was in its interest to carry out the arrangement.

Why This Matters for AI Governance

The Black Sea Initiative operated under extreme conditions: during active military conflict, under comprehensive sanctions, and between belligerents with zero trust. US-China strategic competition over AI, while intense, does not currently involve kinetic warfare or the complete breakdown of diplomatic and economic relations that characterized Russia-Ukraine dynamics in 2023.

So why should a wartime bargaining failure inform our thinking about peacetime AI governance?

Because the Black Sea case reveals what happens when institutional mechanisms confront fundamental divergences in power and interest, and those dynamics are present, even if less extreme, in AI strategic competition. The US and China face structurally similar pressures: asymmetric stakes in maintaining technological leadership, limited ability to verify compliance, weak enforcement mechanisms, and incentives that are likely to shift as capabilities advance. If well-designed verification couldn’t survive Russia’s calculation that defection served its interests better than compliance, we should ask whether AI governance agreements can withstand similar pressures when one party perceives decisive advantage from breaking the deal.

The analogy is not perfect, but the underlying logic is instructive: when defection appears more valuable than cooperation, and enforcement is weak, institutional frameworks struggle to hold, regardless of whether the context is war or strategic rivalry.

Mechanisms of Collapse

Asymmetric incentives shift over time. From the outset, Russia’s unprovoked invasion of Ukraine created an asymmetric situation where a much larger, more populous state attacked a significantly smaller and less populous neighboring state. For Ukraine, the ability to export grain abroad was existential, as it was one of the largest sources of budget financing. For Russia, the arrangement did not have such an existential character, though they wanted to avoid being blamed for global famine in the Global South.

Once the deal was reached, the parties’ incentives started shifting. As global grain prices stabilized, Russia’s incentive changed from “participation” to “obstruction.” They began using the deal to extract concessions regarding their own exports, demanding reconnection of a Russian agricultural bank to SWIFT. At the same time, they created obstacles to implementation, such as slowing down inspections in Istanbul10. Eventually, Russia decided that Ukraine had more to gain from the deal than they did and unilaterally left it.

In AI governance, a similar mismatch could emerge. When one party perceives that unilateral acceleration could deliver decisive strategic advantage, it may face strong incentives to defect, particularly if the leading party uses coordination frameworks to entrench its dominance rather than create genuine parity. Historically, weaker powers have sometimes supported institutional constraints to slow stronger rivals or gain transparency11. But when defection appears to offer a path to advantage in a very crucial race, when racing ahead seems more valuable than rules-based competition, and there is an opportunity to hide noncompliance, institutional commitments become vulnerable. As power dynamics shift, so do incentives to comply with an arrangement.

Exit costs were negligible, and guarantors lacked enforcement power. The Black Sea arrangement did not provide for penalties that could have had a preventive effect. Knowing that departure wouldn’t inflict any significant cost, Russia eventually left the agreement. No sanctions, loss of benefits, or severe economic consequences followed. The reputational cost was also not significant for a country that doesn’t respect international law anyway.

Turkey could act as a mediator, host coordination, and carry out inspections. But it could not compel Russia’s participation when the latter no longer considered it beneficial. Neither could the UN. Neutrality without enforcement power meant guarantors were not in a position to prevent non-compliance.

AI agreements face similar vulnerabilities. What penalties would follow a violation? Who would be in a position to enforce them? No country controls both US and Chinese compute infrastructure, talent pools, or research ecosystems. If there is an institution charged with this function, as long as it lacks tools to enforce penalties and common understanding among its members, compliance will depend on continued willingness, not fear of consequences. The UN Security Council offers a cautionary example: even in cases of grave violations of international order, the institution is paralyzed by veto power granted to countries with diverging interests12.

AI governance faces additional complexities. Grain shipments moved through visible physical chokepoints while AI development takes place in concealable data centers. Missed grain exports could be compensated later while AI capability gains may create advantages that are difficult to neutralise. The grain deal operated on monthly timelines while major AI breakthroughs can happen in days, with detection lagging violations by months. These factors make it not only more difficult to negotiate and reach any deal but also harder to sustain compliance afterward.

Subscribe now

The Nuclear Exception And Its Limits

There were cases in the past where, faced with existential threat, powerful rival states reached understandings that held. During the Cuban Missile Crisis in 1962, President Kennedy and Soviet leader Khrushchev negotiated the removal of missiles from Cuba and Turkey to avoid nuclear escalation13. Similarly, Cold War nuclear treaties such as SALT I (1972)14 and the INF Treaty (1987)15 established limits and verification mechanisms between the U.S. and the Soviet Union. These regimes reduced escalation frequency, established crisis communication channels, and produced decades of strategic stability that avoided direct superpower war - achievements that should not be dismissed lightly.

Yet the conditions that made nuclear governance viable may not transfer to AI. Nuclear deterrence worked because mutual vulnerability was symmetric and immediate. Both superpowers could destroy each other within hours, creating powerful shared incentives to avoid escalation. AI systems pose risks that are fundamentally different: delayed rather than immediate, often asymmetric rather than mutual, and offering potential for decisive unilateral advantage rather than assured mutual destruction.

Present research has drawn crucial distinctions between nuclear and AI arrangements16. Nuclear arsenals were physically observable through satellites and other verification methods. AI development is far harder to verify. Data centers can be disguised, breakthroughs concealed for months, and by the time violations become detectable, irreversible capability gains may have already occurred.

Moreover, nuclear technology matured and stabilised by the 1960s. AI is advancing unpredictably, where a single breakthrough could render agreements obsolete overnight and create overwhelming pressure to defect preemptively.

The reality is that even nuclear weapons don’t prevent all forms of aggression between rival powers. Consider India and Pakistan. Both countries have possessed nuclear weapons since 199817, with India conducting tests on May 11 and 13, 1998, and Pakistan conducting tests on May 28 and 30, 199818. Yet in 1999, just one year after both went nuclear, they fought the Kargil War, a conventional conflict that resulted in 527 Indian soldiers killed and estimates of Pakistani casualties ranging from 400 to over 1,00019. Nuclear deterrence prevented escalation to total war, but it did not prevent war itself.

In my home country of Ukraine, there is often discussion of what might have been had we not abandoned our nuclear warheads under the Budapest Memorandum20: would Russia still have launched a full-scale invasion?

If even nuclear arrangements, which benefited from symmetric vulnerability and immediate catastrophic consequences, could not guarantee peace in all circumstances, what confidence should we have in AI governance agreements that lack these stabilizing features?

Facing Uncomfortable Realities

Some risks may prove difficult to manage through international guarantees alone. This is an uncomfortable possibility, but acknowledging it is an argument for strategic honesty, not fatalism.

International treaties regulating relationships between major AI powers may struggle to hold under pressures similar to those that existed in the Black Sea Grain Initiative. When defection incentives are strong, violations are hard to detect, exit costs are negligible, and enforcement is weak, international agreements face serious challenges.

This should not lead to a fatalistic conclusion, though. There are already alternative approaches to AI safety that acknowledge these limits honestly. Arrangements could be reached between smaller coalitions of genuinely aligned actors, for instance, the US, EU, UK, and allied nations could establish binding commitments on frontier AI development without China’s participation21, accepting this won’t prevent Chinese advances but ensuring allied coordination. Similarly, compute governance through hardware verification and export controls could leverage the concentrated nature of semiconductor supply chains22. These approaches allow control of supply chains rather than relying solely on promises.

What ultimately worked in the case of the Black Sea Initiative was not the mutual agreement but Ukraine’s development of military capabilities. After the deal collapsed, the primary mechanism that made export of Ukrainian grain possible were sea drones23. These autonomous vessels accomplished what neutral third parties, technical verification, and declared commitments could not: they kept the Russian Navy away from the grain routes through credible deterrence, not diplomatic assurances24.

The lesson is stark but important. Ukraine’s grain exports resumed not because Russia honoured its word, but because Ukraine developed the capability to defend its interests independently. Safety, whether of grain shipments or AI development, depends on the ability to protect what matters to you, not on promises from rivals with diverging interests. When the stakes are existential, power and capability matter more than institutional frameworks and verification procedures.

This should inform how we approach transformative AI development. The party that maintained grain exports was not the one that honoured commitments, but the one that built countermeasures its adversary couldn’t overcome. Similarly, AI safety may ultimately depend less on international guarantees than on maintaining strategic capability: technical defences, hardware controls, and coalitions of aligned actors who can enforce standards without depending on adversary compliance. Power and political calculations will determine whether international AI governance succeeds or fails, and betting everything on promises from rivals may prove as ineffective as expecting Russia to honour the grain deal out of goodwill.

Author’s Note on AI Collaboration

This article was written by the author. However, Anthropic's Claude, Google's Gemini, and OpenAI's ChatGPT were utilised as thought partners during the production process. Specifically, the AI provided:

• Editorial Polishing: Refining the prose for flow and professional tone.

• Critical Review: Stress-testing the arguments and identifying potential gaps in the author’s analysis.

• Content Suggestions: Assisting with the technical synthesis of footnotes and structural formatting.

All ideas, final decisions, creative direction, and factual verifications were made by the human author.

Leave a comment