A green and yellow three-wheeler on a street moves past BRICS summit banners bearing images of Indian Prime Minister Narendra Modi, Russian President Vladimir Putin and Chinese President Xi Jinping

BRICS 2026 summit banners bearing images of Indian Prime Minister Narendra Modi, Russian President Vladimir Putin and Chinese President Xi Jinping in New Delhi, India, Sept. 12, 2026. (AP Photo/Mukhtar Khan)

Before India Exports Digital Public Infrastructure, It Should Define It at Home

CHONGQING, CHINA - AUGUST 10: In this photo illustration, the logo of Meta is displayed on a smartphone screen on August 10, 2026 in Chongqing, China. (Photo by VCG/VCG via AP )

Meta Brings Community Notes to LatAm Without Showing What Success Looks Like

The Léon Thévenin near the Buitenratel, Belgium in 2009. Hans Hillewaert/Wikimedia Commons/CC by 4.0

The Léon Thévenin near the Buitenratel, Belgium in 2009. Hans Hillewaert/

The UN’s AI Agenda Must Prioritize Undersea Cables

US President Donald Trump and Chinese President Xi Jinping attend a welcome ceremony at the Great Hall of the People on May 14, 2026 in Beijing, China. (Photo by Alex Wong/Getty Images)

Trump and Xi Should Talk About Worms

A military band conductor with raised arms with a view from below of an ornate ceiling with a red star in the middle

A military band conductor leads during a ceremony to mark the 105th anniversary of the Chinese Communist Party at the Great Hall of the People in Beijing, July 1, 2026. (AP Photo/Ng Han Guan)

US, China Psy-Op Accusations Signal Deep Distrust on AI Regulation

Secretary-General Antonio Guterres speaks during press conference ahead of General Assembly High-level Week 2026 at UN Headquarters in New York, NY on September 16, 2026.(Sipa via AP Images)

Can the United Nations Help Slow the AI Race?

The setting sun reflects off an Anduril Sentry tower in Luna County, New Mexico. EFF/Wikimedia Commons October 2022

The setting sun reflects off an Anduril Sentry tower in Luna County, New Mexico. EFF/

The Influence of AI on Human Decisions in DHS Surveillance

Interior view of House of Commons at Canada Parliament. Ottawa, Canada - December 15, 2025. Shutterstock

Interior view of House of Commons at Canada Parliament. Ottawa, Canada - December 15, 2025. 

Canada Is Not Banning Teenagers from Social Media — It’s Trying Something New

ICE agents stand guard outside Delaney Hall detention center in Newark, New Jersey, on Thursday, May 28, 2026. (Photo by Cristina Matuozzi/Sipa via AP Images)

From Records to Raids: The Questionable Legality of ICE’s Palantir ELITE System

European Commission President Ursula von der Leyen delivers a speech at the annual State of the European Union at the European Parliament in Strasbourg, Wednesday, Sept 16, 2026. CC-BY-4.0: © European Union 2026– Source: EP

Europe Says Its AI Rules Are Enough. AI Agents Are Testing That Claim

United States President Donald Trump steps off from Marine One upon his arrival at the Ellipse just outside of the White House, Saturday, Sept. 19, 2026, in Washington. (AP Photo/Jose Luis Magana)

AI Hacks, a Possible War Crime, and a Near Miss With China Set Stage for Trump-Xi Meeting

The Google sign on the wall behind the reception desk at the Google headquarters in Silicon Valley, August 2023. Shutterstock

The Google sign on the wall behind the reception desk at the Google headquarters in Silicon Valley, August 2023. 

A Deep Dive into the Antitrust Remedies that Spared Google's Ad Tech Business

Online Radicalization Concept with Digital Propaganda, Social Media Manipulation, Extremist Influence, Algorithmic Amplification, Hate Speech, Misinformation, Polarization and Online Safety Risks

Matthias J. Becker on the Launch of Digital Hate Review

Ambiance at the Global Dialogue on AI Governance, Palexpo, Geneva, Switzerland - 07 July 2026.

The Global Digital Compact was adopted at the UN General Assembly in New York City on September 23rd, 2024. 

UN's Global Digital Compact Review Could Sideline the Global South

An official of the Lagos State Independent Electoral Commission (LASIEC) marks a voter's finger with ink in Lagos, Nigeria, on July 12, 2025. (Photo by Adekunle Ajayi/NurPhoto via AP)

When Nigeria’s Election Technology Cannot Recognize the Voter

A Google logo embedded in a cracked, black surface. The cracks radiate outward, giving the appearance that the logo is breaking through or shattering the material. The Google logo retains its signature colors: blue, red, yellow, and green.

Can the EU Still Break Up Google?

Artwork including a giant bust of Elon Musk sits along the route the billionaire's SpaceX Starbase south Texas launch site. (Photo by Laura Brett/Sipa via AP Images)

What Policymakers and the Public Need to Know About the Cult of AI

A bald eagle, signifying the US Department of Justice, next to the Google logo.

The Google Remedies Turn Landmark Verdicts Into a Whimper

CEO of Anthropic Dario Amodei, addresses the gathering at the AI Impact Summit, in New Delhi, India, February 19, 2026. REUTERS/Bhawika Chhabra

Who Should Pace the Frontier? Not Dario Amodei

A traditional Chinese arch bridge over green water surrounded by trees sits in front of large skyscraper buildings.

A traditional Chinese arch bridge sits in front of the 'Huijin International Building' — the building that hosts DeepSeek's headquarters, in Hangzhou, Zhejiang Province, China, on March 18, 2026. (Photo by STR/NurPhoto via AP)

China Is Running a Different AI Race

When United States President Donald Trump and China President Xi Jinping meet in Washington, D.C., on 

, artificial intelligence will loom over the summit as a central arena of strategic competition between the two countries. Ahead of the meeting, US and Chinese officials considered establishing a 

 that threaten national security. AI-enabled adaptive computer worms offer a concrete example of the kind of crisis such a mechanism could help manage. By revising their attack strategies as they spread, such malware could exploit vulnerabilities at scale and place sophisticated cyber capabilities within reach of non-state actors. The summit therefore offers an opportunity for Washington and Beijing to begin building a framework for cooperation against an emerging threat that could endanger both countries.

Computer worms are a type of malware that can self-replicate and autonomously spread across networks. Once deployed, they can move far beyond their initial targets and cause widespread damage. In 2017, Russia deployed a computer worm known as NotPetya against Ukraine. The attack disrupted Ukrainian government agencies and critical infrastructure before spreading beyond the country through the networks of multinational companies operating there. NotPetya compromised 

, making it one of the most destructive cyberattacks in history. NotPetya’s global spread demonstrated the transnational risk posed by computer worms, as an attack intended for a single target or country can inflict damage worldwide.

The damage caused by computer worms could become far more severe as advances in AI give them new capabilities to adapt, evade defenses, and exploit systems as they spread. Traditional worms spread automatically, but their attack logic is largely hard-coded and predetermined before deployment. Attackers must anticipate the vulnerabilities and system configurations a worm will encounter in advance, and its spread can stall when those assumptions prove incorrect. In June, 

 a proof-of-concept adaptive computer worm, powered by a large language model, inside an isolated test network. Unlike traditional worms that rely on predefined attack logic, the worm could reason about the systems it encountered, generate and revise attack strategies, exploit vulnerable machines, and continue self-replicating without human intervention. It could also run on a locally hosted open-weight model, allowing it to operate without relying on a cloud service or commercial AI provider. As the worm spread, compromised machines with sufficient computing power could provide additional resources for reasoning and further attacks.

The geopolitical significance of AI-enabled adaptive worms lies in their potential to enable non-state actors to conduct cyberattacks at a scale previously beyond their reach. By reducing the need for human involvement in each individual compromise, adaptive worms could allow attackers to carry out tailored attacks across far more systems than a team of human operators could manage. AI is separately reducing the human effort required to build such tools. The Palo Alto cybersecurity firm Calif reported that AI helped its researchers develop 

, that could hijack WeChat accounts through unanswered calls in about a week, work that would previously have taken a larger team months. AI-enabled adaptive worms’ ability to run on locally hosted open-weight models could further lower barriers by reducing attackers’ dependence on cloud-based AI providers and limiting the effectiveness of safeguards such as service refusals, account suspensions, or API restrictions. Together, these capabilities could make sophisticated cyber operations more accessible to cybercriminal organizations, terrorist groups, and other non-state actors that previously lacked the resources or expertise required to carry them out.

AI-enabled adaptive worms could outpace existing approaches to cyber incident response. Investigations, attribution, and law-enforcement responses generally occur after an attack has begun and can take months or years. Historically, this delay has been partly manageable because defenders can often contain or disrupt malware while those processes continue. The 

, for example, was curtailed after researchers discovered a kill switch, while botnets can often be disrupted by seizing their command-and-control infrastructure. AI-enabled adaptive worms could make these containment strategies far more difficult. A worm capable of revising its attack strategy as it spreads could continue propagating even after defenders patch the vulnerability it initially exploited. Its ability to operate from compromised machines also makes centralized disruption more difficult, since disabling a cloud service or command-and-control server would not necessarily stop the worm’s proliferation. Once such a worm is spreading autonomously, identifying or arresting the original operator may likewise do little to halt the attack.

The transnational nature of AI-enabled adaptive worms creates a common interest in preventing their deployment and containing their spread. Both the United States and China have already experienced the consequences of large-scale computer worm attacks. During the 2017 WannaCry outbreak, China’s national computer emergency response center reported that 

roughly 18,000 Chinese IP addresses were affected

, with disruptions reaching universities, energy systems, and other critical networks. The US was also affected, with the Department of Health and Human Services reporting significant 

operational disruptions at two large, multi-state hospital systems

. WannaCry demonstrated how self-replicating malware can cross borders and disrupt systems far beyond its initial targets. AI-enabled adaptive worms could magnify this mutual risk, threatening both countries regardless of where an attack originates or whom it initially targets.

AI-enabled adaptive worms could provide an early test of whether limited U.S.-China cooperation on AI governance is possible. Preventing non-state actors from gaining access to the most capable AI models is one area where the two governments have already signaled shared interest. 

During the May 2026 U.S.-China summit in Beijing

, Treasury Secretary Scott Bessent said the two sides would establish a protocol for best practices aimed at preventing non-state actors from exploiting the most powerful AI models. This week, US and Chinese officials 

 a bilateral notification mechanism for serious AI-related incidents that pose national security risks. Such a “

” could allow cybersecurity authorities to quickly exchange information during a major outbreak without waiting for attribution. Trump and Xi could go further by agreeing to limits on the use of AI-enabled adaptive worms against civilian systems and critical infrastructure, while also pledging to investigate and disrupt non-state actors that deploy them. Together, these measures could create a limited framework for preventing and responding to AI-enabled adaptive worms without requiring broader agreement on AI governance.

Reaching even a narrow agreement would require overcoming significant mistrust between Washington and Beijing. AI cyber capabilities are inherently dual-use, since the same capabilities that can help identify vulnerabilities and strengthen cyber defenses can also support offensive operations. Yet the US and China have already shown that broader strategic competition does not preclude agreement on specific AI risks. 

During President Joe Biden and Xi’s November 2024 meeting

, the two leaders affirmed that decisions to use nuclear weapons should remain under human control. The meeting did not resolve broader disagreements over AI or national security, but it demonstrated that the two countries could establish a narrow principle around a single high-consequence AI risk. Cooperation on AI-enabled adaptive worms could extend this approach to the proliferation of advanced cyber capabilities to non-state actors. Such cooperation would not resolve broader disagreements over AI governance, but it could establish another precedent for addressing risks emerging from this technology.

This week’s summit offers a rare opportunity to act before a shared risk becomes a shared crisis. Conventional computer worms have already caused billions of dollars in damage, and AI could make future attacks significantly more destructive and difficult to contain. The threat from AI-enabled adaptive worms is no longer theoretical, but such systems have not yet appeared at scale in real-world attacks. That gives Washington and Beijing a narrow window to establish guardrails and put the proposed AI emergency “hotline” into operation before it is urgently needed. The first major AI-enabled worm attack should test US-China preparedness, not mark the beginning of US-China cooperation.


India provides software to countries that want to build their own digital public infrastructure (DPI), but has yet to answer important questions about DPI at home.

In the past months, India has 

 DPI cooperation agreements for “population-scale digital solutions” with at least 24 countries. At the 

 on Aug. 20, the Minister of State for Telecom 

 that India would share modular, open source digital public infrastructure (DPI) frameworks that its 10 other member states could adapt.

In practice, this means a government can 

 its own biometric identity database on the open source platform underlying Aadhaar, which has 

 nearly 1.4 billion enrollments, or develop an instant-payments system on the Unified Payments Interface (UPI), which 

 ten years in August and carries the world's largest volume of real-time retail payments.

There is nonetheless a statutory gap. India has defined DPI in policy documents, but no statute determines what such a digital system precisely is — or what safeguards attach to one. The term carries weight in policy, procurement and diplomacy, but none in law.

 adopted in New Delhi under India's presidency in 2023, DPI is defined as digital systems built by public and private actors on open standards, delivering services at societal scale. The declaration also states that such infrastructure should be accountable and respectful of personal data, privacy and intellectual property rights. Yet in India, there are no legal obligations for DPI operators to protect rights, nor is there a definition of a DPI system.

Last year, the chief executive of Digi Yatra, an industry-led initiative to introduce paperless airport check-in and security using facial recognition, 

 that the platform aims to secure “DPI status” to take its technology beyond aviation into hospitality, office access and examinations. What is being sought, and from whom, went unstated.

, two kinds of DPI. Statutory ones such as Aadhaar rest on legislation and a statutory governance architecture. Hybrid ones — UPI, Open Network for Digital Commerce, Digi Yatra — are run by Section 8 companies operating with state endorsement, without any dedicated DPI statute. They perform functions of enormous public significance with substantial state backing, yet, not being state instrumentalities, 

 the Right to Information Act, 2005, and carry none of the accountability that attaches to a statutory authority.

The consequences are visible in the largest hybrid system. The National Payments Corporation of India (NPCI) sets UPI's rules, standards and transaction limits, while operating its own products on the same network. A 

 commissioned by the Ministry of Finance in 2016 found that just ten large banks held roughly three-quarters of NPCI's equity, and recommended more diffuse ownership and clear separation between infrastructure and regulatory functions. Because there was no obligation, the recommendation went unimplemented: NPCI still promotes its own state-owned mobile payment app, BHIM, competing against the very apps it regulates.

That is not merely an appearance problem. NCPI decides who is admitted to the network, on what technical and operational terms, with direct bearing on the competitors of its own product. Yet no law classifies it as a regulator. Nor is it an ordinary private company, since it performs a public function under the aegis of the Reserve Bank. It therefore falls outside the purview of the Right to Information Act, under which it could be held accountable. That gap is not a mere oversight in the design of NPCI. It follows from the fact that nothing in Indian law clearly defines what a DPI is nor what legal obligations and statutory responsibilities must attach to one.

Digi Yatra exemplifies the problem even more sharply. The Digi Yatra Foundation manages the Digi Yatra ecosystem. The Airports Authority of India, a statutory authority 

 the Union Ministry of Civil Aviation, is its largest single shareholder, 

Yet its governance has been opaque. The government maintained that passenger data remained on the traveller's device and was shared only with the airports. However, 

 on the original Digi Yatra application in 2024 found that data was allegedly also being transmitted to Amazon Web Services (AWS) infrastructure associated with Dataevolve, the private vendor that built the system. Because no statute required a public, independent impact assessment, that divergence between the stated architecture and the operating one went unexamined until the system had already reached millions of travellers.

Currently, India’s Digital Personal Data Protection Act, 2023 (DPDP Act) does not close that gap. It governs how personal data is handled once a system is running: notice, consent, security, and a fiduciary's answerability for its processor. It does not require anyone to establish, before deployment, that a system operates the way its sponsors describe.

For ordinary data fiduciaries, the DPDP Act does not mandate a pre-deployment impact assessment or independent audit. Those obligations attach only to entities separately notified as Significant Data Fiduciaries (SDFs), which must undertake annual DPIAs and audits.

Defining DPI matters because, at present, a system does not attract stricter safeguards simply because it functions as DPI. Its operator must still be separately notified as an SDF. If that does not happen, even a system operating at population scale may fall outside requirements such as mandatory impact assessments and independent audits.

These questions have now reached the courts. In 

Digi Yatra Foundation v. Data Evolve Solutions

 whether the Foundation or its former vendor retains IP rights over the underlying software.

The dispute exposes a significant governance question: who controls the infrastructure that processes the personal data of over 

? The court intervened in March 2024, restraining the Dataevolve from using or copying the data, while directing the handover of server access, source code, domain certificates and cloud credentials to Digi Yatra. That the court needed to spell out what control of the platform entails, is itself the point. These data protections came as interim relief in a commercial dispute, rather than from any standard attached to Digi Yatra as public digital infrastructure. The suit still remains pending, though it was framed for trial in Oct. 2025.

Accountability, human rights, personal data and intellectual property are among the various

 India endorsed together with the G20 in 2023. They are also, precisely, what this suit is about, and none are on track to be settled by anything other than a contractual dispute between a Digi Yatra and its vendor. We consider this gap as a serious risk.

A proper legal status for DPI could require an operator to retain control over critical infrastructure and data, and prescribe minimum terms for contracts with private vendors, including access to source code and cloud credentials, audit rights, restrictions on use of personal data, and transition obligations when a vendor exits.

Laws do this elsewhere already. In banking, the Reserve Bank of India (RBI) 

 outsource a critical technology entirely to a contractor. Regulated entities must retain oversight, audit and access rights, and plan for the return or transfer of data and systems when the relationship ends. A DPI framework could do the same. It could also attach transparency obligations to DPI operators even where, as with the Digi Yatra Foundation, their present corporate form as Section 8 companies, excludes them from the Right to Information Act.

 envisage DPI and AI converging, with government applications such as 

 for language and translations built on shared rails, and national compute moving past 58,000 GPUs. Once AI agents transact on these systems, questions of liability and data provenance will multiply and will be hard to settle for a category with no legal boundaries. AI compounds the problem by involving more actors, more uses of the same data, and potentially more competing claims over the same systems.

 is that these issues matter to the expanding list of India’s DPI partner countries. Bangladesh 

 its UPI-inspired Binimoy platform in 2025 and is rebuilding with open-source 

 software. Sri Lanka holds an Indian grant of about $35 million for its 

 how to fund the rest of the cost. Nepal has 

 UPI for cross-border transfers, but has not decided on a digital identity program. What remains open in each case is institutional as well as technical: how the system is governed, who answers for it, and what it costs.

An Indian statute would not decide those questions for another sovereign, but it would supply a reference position. A recipient government evaluating the Indian stack against another vendor compares the software architecture as well as the terms that come with it. At present India can point to no baseline: no test for what qualifies as DPI, no obligations that attach on qualification, no account of where ownership of the infrastructure, control of the data and rights in the software are meant to sit. Each partner therefore negotiates those terms from scratch, as Sri Lanka is now doing, and as Bangladesh did before abandoning the model. A definition would give India something to offer beyond the software, namely a shortcut to its own governance model.

For a statutory definition to offer clarity to hybrid platforms, it needs to do only two things. 

, set a threshold: population scale, open standards, interoperability, and a function on which access to a service or entitlement effectively depends. 

, provide for legal obligations and statutory responsibilities: publish impact assessments, enforce purpose limitation, extend RTI Act transparency, and provide independent grievance redress.

The timing is favourable. Substantive obligations under the Digital Personal Data Protection Act, 2023 become 

 in May 2027, and the entities they would apply to are mapping their systems and data governance structures now A definition framed now would be absorbed into work already underway rather than retrofitted later, and it would add what the Act leaves out: obligations that attach to a system because of 

 it is, not merely because of what it processes.

India has built infrastructure the rest of the world wants to emulate. The rulebook of data protections that goes with it would be a valuable contribution. A country that defines DPI in law exports a 

 as well as the software that billions of people will depend on worldwide.

This week, world leaders convene in New York for the 81st session of the United Nations General Assembly (UNGA). At the general debate, leaders are expected to make statements that address "

complex and interconnected global challenges

," and, as in recent years, there will be 

 on artificial intelligence. This year, the AI conversation has been heightened by reports of AI agents

. In July, OpenAI models escaped a controlled test environment, executed commands without human direction, and breached the systems of the AI firm Hugging Face in what is believed to be the 

The incident has stimulated the debate on AI governance, with at least three 

 emerging: those who want regulation to slow AI development, those who say full steam ahead and let the 

 self-regulate, and a middle camp that favors progress with guardrails. This debate dominates the media and policy circles and will, no doubt, carry into UNGA this week, where the United States is 

 an event on AI risks on the sidelines of a UN Security Council meeting on AI and security. This focus on the advancement of AI is necessary; however, it risks overshadowing other important conversations on AI governance, including the protection of the infrastructure that drives AI — undersea cables chief among them.

Why undersea cables are an AI governance priority

Undersea cables have become a concern for many countries, especially in the AI era. Cables carry an estimated 

 of international data traffic and are the rails on which AI runs: training data, inference traffic, and cloud services. They 

 across continents and carry the cross-border traffic AI agents depend on to execute commands. Unlike traditional internet traffic, AI 

 on massive and continuous flows of data between hyperscale data centers, cloud providers, and enterprise customers.

These cables, however, introduce another order of risk to AI, a vulnerability inherent in their architecture — they break. When they do, international telecommunications data stops flowing, causing internet outages. For AI, breaks mean disruptions to model training and agentic AI services, making undersea cable cuts an interconnected AI conversation.

 Africa, and elsewhere — often at significant financial cost. In 2025, for example, cable cuts in the Red Sea degraded internet connectivity across the Gulf for days, causing 

 cable damage is traced to human activity, fishing in particular, though natural events such as underwater landslides also cause outages. Suspected deliberate damage to undersea cables has risen in recent years, raising concern especially in 

. How quickly a region recovers from a break depends on the density of its network, or redundancy. Across much of the Global North, overlapping undersea and terrestrial routes absorb a fault, and most users never notice it. In regions where cable networks are still thin, such as small island developing states and developing regions like Africa and parts of the Asia-Pacific, a single fault can cause outages lasting days and sometimes weeks.

, when four submarine cables serving the West African coast (the West Africa Cable System, Africa Coast to Europe, MainOne, and SAT-3) were damaged by an underwater landslide off Côte d'Ivoire. Thirteen countries lost connectivity, some of them almost entirely. The costs across the region were steep; in Nigeria alone, losses were estimated at 

 over four days. The shortage of repair vessels in the region made it worse: the first, the 

, Africa’s only permanently stationed repair vessel, arrived weeks after the incident.

Cable vulnerabilities are not confined to developing regions. The Trump administration has significantly intensified its focus on undersea cables, viewing them as 

 of an ongoing geopolitical race with China. Security analysts have warned that US cables are 

, with adversaries already investing in the means to attack American undersea infrastructure, which they say could be catastrophic for the US economy and political stability. 

 has stepped up its undersea cable security agenda, with 

 following multiple incidents, including cable cuts in the Baltic Sea, some of them linked to 

This makes the protection and resilience of undersea telecommunications cables a global issue that UNGA 81 cannot treat as optional. Alongside the emerging threat of cyberattacks by AI agents, a more mundane AI kill switch exists: cable cuts, a real risk with 

 consequences. The General Assembly must therefore balance its digital agenda to give all these conversations equal prominence.

Governing AI by building the resilience of undersea cables

The vulnerabilities in undersea cables have not gone unnoticed by the UN and world leaders. To build the cables' resilience, the UN agency for technology, the International Telecommunication Union (ITU), in collaboration with the International Cable Protection Committee, established the 

International Advisory Body on Submarine Cable Resilience

, a global convening charged with strengthening cooperation between stakeholders and identifying practical best practices for the deployment, operation, protection, and repair of submarine cables to enhance resilience at national, regional, and global levels. On July 10, the Advisory Body approved its 

 on undersea cable resilience, a landmark document that sets out preventive and rapid-response measures to reduce the risk and impact of cable cuts, including those that disrupt AI. Yet the report passed quietly, overshadowed by other UN events of the week, including the first 

UN-mandated Global Dialogue on AI Governance

World Summit on the Information Society Forum 2026

The ITU report offers a global reference that countries can use to develop harmonized approaches to undersea cable resilience. It sets out 39 recommendations across three focus areas: timely deployment and repair, covering streamlined permitting, stronger coordination, and improved access to repair resources; risk identification, monitoring, and mitigation, examining the evolving range of risks and the importance of enhanced monitoring, reporting, and coordination mechanisms; and fostering geographic diversity, addressing structural and investment challenges and identifying pathways to strengthen redundancy and expand resilient connectivity, particularly for vulnerable regions.

The recommendations aim to make undersea cables and the services they provide more resilient and achieving them requires global effort. However, like other UN reports, the recommendations are non-binding, so adoption rests with governments and industry to keep championing the issue and pressing the conversation forward, especially for regions most exposed to cable vulnerabilities. Africa is one of them.

The continent's submarine cable architecture is underserved, both by cable networks and by repair vessels, leaving the region exposed. Africa’s sole repair ship (the Léon Thévenin) is based in South Africa, meaning much of Africa's cable network runs through waters with no nearby maintenance capacity. Should a fault occur near Dakar (on Africa’s Atlantic seaboard) while that vessel is working off Mombasa on the continent’s eastern seaboard, the transit alone would exceed 10,000 kilometers, a distance comparable to a transoceanic route such as Shanghai to Los Angeles. The 2024 West African outage showed what this means in practice: Africa's cable repair ship arrived weeks later as it was already commissioned elsewhere in the continent and even after it arrived on site it could only attend to one of the four damaged cables, while the other three 

In the AI age there is an imperative for African countries to adopt the recommendations of the report especially those on timely deployment and repair and geographic diversity. 

The United Nations Convention on the Law of the Sea

 (UNCLOS), the foundational legal framework for undersea cables, depends on countries to give its protections effect through national law, yet many African countries lack legal frameworks to protect and manage cables. The ITU report recommended that countries designate undersea cables as critical infrastructure and implement UNCLOS Article 113, which requires states to make damaging a cable a punishable offense. This will require national legislation and a model already exists in Nigeria’s 

Critical National Information Infrastructure Order, 2024

. It is now for other African countries with cable landings to adapt similar measures, alongside the report's other recommendations.

At UNGA 81, the seemingly novel risks of AI agents will most likely steal the limelight. But a robust AI dialogue must also confront the vulnerability of the undersea cables AI runs on. Their protection and resilience is an essential AI agenda, and in New York it must be elevated as such.

The European Union says its AI rules already give regulators the tools to deal with the risks posed by increasingly capable AI models. But as AI systems 

repeatedly cross the boundaries of controlled testing

 and reach real-world systems, legal experts are questioning how far those powers actually extend.

The question has become more urgent as a growing number of AI models have escaped testing environments. Google 

 that a Gemini AI model accessed the systems of three real companies during a cybersecurity test, after finding information online and guessing credentials it believed were within the scope of the evaluation. 

The incidents have also fueled a call to slow the pace of AI development. Anthropic CEO Dario Amodei published an essay titled 

. The text argues that a “pause” is necessary so safety research can catch up. OpenAI CEO Sam Altman, Google DeepMind’s Demis Hassabis and Elon Musk 

For Hamish Hobbs, Director of AI Policy at the Centre for Long-Term Resilience, “recent incidents have made it clear that current safeguards against AI threats are entirely inadequate.” “Right now, even staff at AI labs feel powerless to control the pace of change in the face of escalating risks,” he told Tech Policy Press.

The urgency has reached Brussels. In her 

 last week, European Union Commission President Ursula von der Leyen said she would bring together leading AI labs to discuss how to “pace” the technology.

But Brussels has pushed back on the idea that new safeguards are needed. Asked about Amodei’s proposal during a midday briefing last week, a Commission spokesperson said the EU already has “everything in place” to enforce security safeguards.

The European Commission told Tech Policy Press that under the AI Act providers of advanced models must assess and mitigate systemic risks, including potential loss of control, and such obligations apply “across the entire lifecycle of the model, from the start of its large pre-training run until its retirement.”

Can the AI Act reach models still in testing?

 allow the European regulators to ask providers to restrict a model’s availability on the EU market, withdraw it or recall it. But experts disagree over how those powers can be used in the kinds of testing incidents now making headlines.

It remains unclear how those powers apply when a model has not been placed on the market, but an accident occurring during testing affects a real system. The Commission has already sent its 

, focusing on how providers protect their models from security threats, allow independent external testing, and carry out post-market monitoring.

Brussels has separately sought more transparency from providers that have not published summaries about the data used to train their models. However, that’s just “a necessary first step," but "information requests on their own are not enough to enforce the AI Act,” 

 Risto Uuk, Head of European Policy and Research of Future of Life Institute and Co-Founder of the KU Leuven AI Safety Lab.

 that OpenAI failed to submit a report, required under the AI Act, regarding an accident that happened in May when its models escaped testing grounds and interacted with RubyGems.

Brando Benifei, a member of the European Parliament and lead negotiator on the AI Act, also believes Brussels has the legal powers, but argues the AI Office needs more support to act. “The AI Office can obtain model access, run independent evaluations, require mitigation, and ultimately restrict or recall dangerous models placed on the EU market. The Commission must give the Office the political backing, resources, and technical expertise to act immediately,” he told Tech Policy Press.

Rather than banning research, Benifei argues that Europe needs clearer rules “to disincentivize corporate irresponsibility,” backed by enforcement and dissuasive fines.

Harshvardhan Pandit, a researcher at the 

 in Trinity College Dublin, sees market restriction as one of the strongest tools in the AI Act. But he also points to limitations: “Providers can say they have addressed the issues, or that they have shut down a model and are using a different one,” he told Tech Policy Press. “The more important question should be what powers can be used to stop such incidents happening in the future.”

One option, Pandit said, is to scrutinize whether companies are following the commitments agreed in the 

. OpenAI and Anthropic, both signatories, have committed to 

 including reassessing risks after serious incidents, reporting cyber incidents and introducing new safeguards when risks are no longer acceptable.

Pandit said regulators should clarify what mitigation measures could be required, potentially including restrictions on internet access until safety problems are addressed.

Gianmarco Gori, a guest professor and postdoctoral researcher at the Law, Science, Technology and Society at Vrije Universiteit Brussel’s, said there is no clear consensus on how the AI Office’s powers apply to models still in testing. The fact that these “powers can be exercised, especially in cases of models not yet released, raises complex interpretive questions,” he said.

The expert described the AI Act as a “product legislation,” built around concepts such as “placing on the market” and “putting into service.” That creates a question for incidents involving models that have not formally entered the market but have nonetheless interacted with real systems.

, under which a manufacturer can argue that a product left control against its will. But with AI, that raises the question of whether saying “I didn’t want this to happen” is enough or if regulators should assess what the provider actually did to prevent it.

The incentives facing AI companies also complicate the picture. As companies compete to develop increasingly capable systems, Gori said, the pressure to move quickly remains strong. “What happened today, tomorrow is already old,” he said.

Nevertheless, the AI Act is not the only law triggered by the incidents. According to Gori, “specific characteristics of the case may trigger the competence of different regulators, including data protection, cybersecurity, and law enforcement authorities.” Responsibility can also depend on how many actors sit between the model and the incident.

Who is responsible when AI systems cross the line?

For Pandit responsibility can fall on several different actors: the developer of a model, the person or company that turns it into an agentic system, and the deployer that gives it access to the internet, credentials, code or other tools.“If you are doing it all yourself, then you have to fulfill all of those [responsibilities] by yourself,” he said.

Latest

Podcast

Recent

Our content delivered to your inbox

Thank you!

Around the globe