Digital on Day 2 of UNGA80: Calls for digital inclusion, responsible AI, and collective security

 Lighting, Stage, Purple, Electronics, Screen, Computer Hardware, Hardware, Monitor, Light, Urban, Indoors, Club

Welcome to the second daily report from the General Debate at the 80th session of the UN General Debate (UNGA80). Our daily hybrid AI–human reports bring you a concise overview of how world leaders are framing the digital future.

Day 2 debates underscored a need to align rapid technological change with global governance, with countries calling for stronger international cooperation and responsible approaches to the development and use of technology. Delegations emphasised that digital technologies must serve humanity – advancing development, human rights, and democracy – while also warning of the growing security risks posed by AI misuse, disinformation, hybrid warfare, and cyber threats. Alongside some calls for rules and ethical standards, many highlighted the importance of inclusion, investment in digital infrastructure, and ensuring that all states can share in the benefits of the digital age.

To keep the highlights clear and accessible, we leave them in bullet points — capturing the key themes and voices as they emerge.


Global digital governance and cooperation

  • Technological disruptions are currently outpacing governance. (Kenya)
  • The digital age must be guided by international cooperation, ethical standards, and respect for human rights, with technology placed at the service of humanity. (Albania)
  • The world needs a strong and effective UN system capable of responding to the rapid evolution of new technologies. (Czechia) A renewed UN can strengthen digital security and international cooperation with ethical and inclusive principles that support freedom of expression. (Panama)
  • Albania is co-leading with Kenya the review process of the World Summit on the Information Society (WSIS) and will work to ensure a successful outcome. (Albania)
  • International Geneva can make a unique contribution to the attainment of global goals, leveraging its expertise in humanity and innovation as a centre for reflection, discussion, and concerted action. (Switzerland)

Artificial intelligence

Responsible AI (governance)

  • AI must serve human dignity, development, and human rights, and not the other way around. (Estonia)
  • AI governance is seen as one of three significant global challenges facing the international community, along with nuclear weapons and the triple planetary crisis. (Costa Rica)
  • Governments should act swiftly to create regulations that make AI safer and more beneficial for people. Focus should be placed on developing AI  responsibly, not halting progress. (Latvia)
  • A responsible approach from all international institutions, the private sector, and governments is needed to steer the AI revolution. (Slovakia)
  • Regulations, ethical standards, and governance mechanisms are urgently needed in the AI space, to address issues of equity and access. (Guyana)
  • A global standard is called for to ensure the use of AI is transparent, fair, and respects ethical boundaries, without substituting for human judgment or responsibility. (Namibia)
  • The UN General Assembly’s decision to establish two global AI governance mechanisms – the independent international scientific panel and a global dialogue on AI governance –  is welcomed. (Guyana, Costa Rica)

AI for development and growth

  • AI can accelerate progress on the 2030 Agenda if directed towards a fair and equitable digital transformation. (Spain) It can strengthen national economies and collective efforts for development, optimising resources, accelerating medical research, and democratising access to knowledge. (Costa Rica) AI can also promote economic growth, drive scientific progress and innovation, improve healthcare, and make education more accessible. (Latvia) 
  • AI and digitisation can accelerate the demand for energy. (Guyana)
  • Investment is needed in new technologies and artificial intelligence to help developing countries transition to a more prosperous future. (Congo)
  • AI must stand for ‘Africa included‘. (Nigeria)
  • An AI hub for sustainable development is being opened, involving hundreds of African startups in the development of artificial intelligence. (Italy)
  • A neutral sovereign artificial intelligence zone has been proposed. (Sri Lanka)
  • Guyana is establishing an AI hyperscale data centre which will help accelerate digitalisation and improve competitiveness. (Guyana)
  • Equipping citizens with the skills to use AI wisely and responsibly is essential. Estonia is implementing a new ‘Artificial Intelligence Leap’ to provide the best technological tools to students and teachers to maintain a comparative edge in education. (Estonia)

Digital tech, peace and security 

  • Concerns were raised about the impact of drones – with or without AI – on peace and security. The proliferation of drones available to countries with limited resources or non-state actors presents a rapidly evolving security threat, having increased the lethality and changed the economics of war. (Croatia, Latvia, Ukraine)
  • Acts of hybrid warfare include disinformation campaigns, attempts to undermine public trust, cyberattacks, and acts of sabotage carried out by mercenaries recruited online. (Czechia) Damage to undersea cables and GPS jamming are also part of a growing wave of hybrid attacks. (Latvia)
  • Emerging threats such as cyberattacks, hybrid attacks, and the misuse of AI (for instance to spread disinformation or enable attacks on critical infrastructure) challenge international peace, security, and stability. Countering these requires resilience and increased cooperation. (Latvia, Costa Rica)
  • Technologies like AI, cyber capabilities, space technology and robotics can strengthen defences, but can also be misused by hostile actors. Security needs to be rethought, nationally and globally. Rules, safeguards, and cooperation must keep pace with innovation in technologies, to ensure that they can contribute to resilience and stability. The UN must evolve to be able to effectively address such complex challenges. (Croatia, Cyprus)
  • There is an urgent need for global rules on how AI can be used in weapons, comparable in urgency to preventing nuclear weapons proliferation. (Ukraine)
  • Military automation, enabled by AI, challenges the ability to maintain meaningful human control over life-or-death decisions without adequate regulatory frameworks. The conclusion of a legally binding instrument before 2026 is urged to establish prohibitions and regulations for autonomous weapons systems capable of identifying, selecting, and attacking targets without meaningful human control, stressing that no algorithm should make life or death decisions. (Costa Rica)
  • The arms race is resuming, including in cyberspace. (Senegal) Cybercrime and cyber terrorism are emerging challenges. (Guyana)

Human rights in the digital space

  • Safeguarding digital rights and advancing media freedom are critical for advancing democracy and protecting international law-based multilateral world order. (Estonia)
  • It is proposed to establish a global charter for digital governance and ethical AI to protect human rights in the digital sphere. (Central African Republic)

Disinformation and misinformation

  • Concern was expressed about an emerging generation that grows cynical because it believes nothing and trusts less, due to the rapid advancement of technology. (Nigeria)
  • The ‘pandemic’ of misinformation and disinformation is an emerging challenge. (Guyana)
  • The proliferation of misinformation, particularly via digital platforms, has fuelled distrust between countries, targeting elections, trade negotiations, and public sentiment. (Serbia)
  • Disinformation, which gains even greater volume in digital environments, is eroding public trust and is part of the challenges testing the principles of the UN Charter and the UN’s authority. (Dominican Republic; Sierra Leone)
  • Autocracies are deploying new technology to undermine trust in democracy, institutions, and each other. (Australia)

Digital technologies for development

Digital inclusion and access

  • Ensuring that every person and country benefits from the opportunities of the
    digital age is a global challenge. The international community must work together to close the digital gap between states that can and cannot benefit from digital tech and AI as development tools. (Sri Lanka)
  • There is a need for a new dialogue to promote a level of access to technology that allows emerging economies to more quickly close the wealth and knowledge gap. (Nigeria)
  • The digital divide must be closed. (Costa Rica, Nigeria) Advancing digital inclusion and the digital transition is essential for states to meet development goals. (Comoros, Kiribati) 
  • A dedicated initiative is advocated for, bringing together researchers, the private sector, government, and communities to close the digital divide. (Nigeria)
  • Investments are made in digital transformation and the digital economy to foster inclusion and innovation, and ensure no one is left behind. (Albania, Sierra Leone)

Digital public infrastructure and services

  • Digital solutions are vital for overcoming challenges from geographical isolation and limited economies of scale, and are key to enhancing public services, education, commerce, and climate resilience. (Kiribati)
  • The GovStack initiative, co-founded by Estonia in collaboration with the International Telecommunications Union and Germany, provides governments with a
    digital public infrastructure toolbox aimed at modernising digital services by creating a modular, open-source, and scalable framework. (Estonia)
  • Digitalisation is a part of the commitment to sustainable development and the 2030 Agenda goals. (Serbia)
  • Digital democracy is a national aim. (Sri Lanka)

Technology transfers, trade, and critical minerals

  • Many countries need technology transfers and capacity building (Guatemala), and developed countries must honour their commitments in these areas. (Sierra Leone) Solidarity, translated into technology transfers and other measures, is needed. (Dominican Republic)
  • The world urgently needs supply chains that are more reliable, diversified, and resilient. (Paraguay)
  • Allowing critical infrastructure to depend on authoritarian regimes is gambling with both the economy and democracy. (Paraguay)
  • Africa has an abundance of critical minerals that will drive the technologies of the future. Investments in the exploration, development, and processing of these minerals in Africa will diversify supply to the international market and help shape the architecture for peace and prosperity. Countries that host minerals must benefit from them through investment, partnership, local processing, and jobs. (Nigeria)

For other topics discussed, head over to our dedicated UNGA80 page, where you can explore more insight from the General Debate.

Diplo NEWS25 Insta UNGA
The General Debate at the 80th session of the UN General Assembly brings together high-level representatives from across the globe to discuss the most pressing issues of our time. The session took place against the backdrop of the UN’s 80th anniversary, serving as a moment for both reflection and a forward-looking assessment of the organisation’s role and relevance.

New Stargate sites create jobs and boost AI capacity across the US

OpenAI, Oracle, and SoftBank are expanding their Stargate AI infrastructure with five new US data centre sites. The addition brings nearly 7 gigawatts of capacity and $400 billion in investment, putting the partners on track to meet the $500 billion, 10-gigawatt commitment by 2025.

Three of the new sites- located in Shackelford County, Texas; Doña Ana County, New Mexico; and a forthcoming Midwest location, are expected to deliver over 5.5 gigawatts of capacity. These developments are expected to create over 25,000 onsite jobs and tens of thousands more nationwide.

A potential 600-megawatt expansion near the flagship site in Abilene, Texas, is also under consideration.

The remaining two sites, in Lordstown, Ohio, and Milam County, Texas, will scale to 1.5 gigawatts over 18 months. SoftBank and SB Energy are providing advanced design and infrastructure to enable faster, more scalable, and cost-efficient AI compute.

The new sites follow a rigorous nationwide selection process involving over 300 proposals from more than 30 states. Early workloads at the Abilene flagship site are already advancing next-generation AI research, supported by Oracle Cloud Infrastructure and NVIDIA GB200 racks.

The expansion underscores the partners’ commitment to building the physical infrastructure necessary for AI breakthroughs and long-term US leadership in AI.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot

Digital on Day 1 of UNGA80

 Lighting, Stage, Purple, Electronics, Screen, Computer Hardware, Hardware, Monitor, Light, Urban, Indoors, Club

Welcome to the first daily report from the General Debate at the 80th session of the UN General Debate (UNGA80). Our daily hybrid AI–human reports bring you a concise overview of how world leaders are framing the digital future.

Day 1 debates circled around a central message: technology must remain a servant of humanity, not its master. From calls to ensure AI benefits all societies and to build universal guardrails for its responsible use, to concerns over cybercrime, disinformation, and the governance of critical minerals, delegations stressed the urgent need for cooperation, inclusivity, and safeguards.

While opportunities for innovation, development, and peace were highlighted, speakers warned that without global frameworks, the same technologies could deepen divides, fuel insecurity, and erode human dignity.

To keep the highlights clear and accessible, we leave them in bullet points — capturing the key themes and voices as they emerge.


Tech for humanity and common good & global cooperation

  • Technology must be put at the service of humanity. It must be our servant, not our master. (UN Secretary-General)
  • The use of technology and global connectivity is too often twisted by cynical leaders and warmongering regimes, but can be harnessed for the common good. (Slovenia)
  • A vision of AI for all is needed to ensure that tech advancements contribute to the universal values of humanity. (Republic of Korea)
  • Africa must play an active role in defining international roles and standards and ensuring that technology is at the service of humanity. (Mozambique)
  • The international community must ensure that technology lifts up humanity and no country is locked out of the digital future. (UN Secretary-General)
  • Peak technology is picking up pace, opening horizons of opportunity but paving the way for dangerous forces because they are not regulated. New risks are posed by AI, cyber, space and quantum technologies, and while common frameworks exist, they have been weakened or outpaced. Existing rules and institutions need to be consolidated, and frameworks for peace need to be built. (France)

Artificial intelligence

AI inclusion and capacity building

  • AI capacity gaps must be closed. All countries and societies must be able to use, design and develop AI, and benefit from the opportunities the technology offers. (Türkiye, Kazakhstan, Uzbekistan, UN Secretary-General) )
  • AI technologies should be used for the benefit of humanity, not as a new tool of domination. The UN Technology Bank for the Least Developed Countries could play a critical role in closing the digital and technological gap. (Türkiye)
  • A new international cooperation mechanism is proposed to facilitate the exchange of practical solutions and models of AI in healthcare, education, and culture. (Uzbekistan)
  • Not taking advantage of AI means wasting economic opportunities. Countries need to adapt to the challenges imposed by the need to use AI responsibly. (Morocco)

Responsible AI (governance)

  • The advancement of AI is outpacing regulation and responsibility, with its control concentrated in a few hands. (UN Secretary-General) There is a need for universal guardrails, common standards, and ethical norms to ensure transparency, safety, accountability, fairness, and the protection of individual rights in its deployment. The UN’s recent steps to establish an international scientific panel and an annual global dialogue on AI governance are supported. (UN Secretary-General, Kazakhstan)
  • Commitment was expressed to building multilateral governance to mitigate the risks of AI, in line with the Global Digital Compact. (Brazil)
  • AI could lead to a dystopia of deepening polarisation, inequality, and human rights abuses if not proactively managed. It can also be a driving force for innovation, prosperity, and direct democracy. (Republic of Korea)
  • Artificial intelligence poses new challenges to human dignity, justice, and labor, with risks of exclusion, social manipulation, and militarization through autonomous weapons. Addressing them requires understanding how AI works and having robust safeguards in place. (Mozambique)

Cybersecurity and cybercrime

  • Digital technologies come with new security threats, in particular cybercrime. Cybersecurity must be an important component of collective security. (Tajikistan)
  • Viet Nam looks forward to the signing ceremony of the UN Convention Against Cybercrime. (Viet Nam)

Digital technology, peace and security

  • There are risks associated with new technologies, from biotech to autonomous weapons. There is also a rise of tools for mass surveillance and control, which can intensify the race for critical minerals and potentially spark instability. (UN Secretary-General)
  • The US will pioneer an AI verification system to enforce the biological weapons convention. (United States)
  • Digital, space and AI technologies should be used as forces for peace, not tools for domination. (Portugal)
  • The use of ICTs to harm peace, security and sustainable development needs to be prevented. (Turkmenistan)

Human rights in the digital space

  • Technology must serve humanity and be a force for good. It must promote human rights, human dignity, and human agency. (UN Secretary-General) 
  • Regulating digital platforms does not mean restraining freedom of expression, but ensuring that what is illegal offline is also illegal online. (Brazil)

Disinformation and misinformation

  • Digital platforms offer possibilities for people to come together, but they have also been used for sowing intolerance, misogyny, xenophobia, and misinformation, necessitating government regulation to protect the vulnerable. (Brazil)
  • The rise of tools for mass disruption and mass social control is a concern. ( UN Secretary-General)
  • There’s a growing challenge of disinformation being used to undermine democratic institutions and destabilize societies. The international community needs to defend truth as a supreme value. (Lithuania)

Digital inclusion and tech for development

  • To bridge the digital and technological divides is central to building resilient societies. (Portugal)
  • It is important to prevent inequalities in digital development and the use of artificial intelligence between countries. (Uzbekistan)
  • Digital transformation must be balanced, reflect the realities and legitimate interests of all states, and be free from politicization and bias. A proposal will be made to establish a world platform on digital integration. (Turkmenistan)
  • There is a need for a technological and a climate diplomacy that can regulate risks and democratise benefits through genuine transfer and sharing of technology and knowledge, so that technology is a factor of inclusive development. (Mozambique)
  • Sustainable development models need to be based on digital and green transition. For this, countries must invest in R&D, train human resources, develop green infrastructure, and formulate national plans, while developed countries must take responsibility in sharing and transferring technology to developing and underdeveloped countries. (Viet Nam)
  • Nations which benefited the most from industrial and economic development in the past should support developing countries through measures such as technology transfers and adequate financing. (Angola)

Critical minerals

  • Robust regulations need to balance responsible mineral extraction with effective environmental protection. (Nauru)
  • Rich countries are demanding greater access to resources and technology. The race for critical minerals cannot repeat the predatory and asymmetrical logic of past centuries. (Brazil)
  • Critical minerals need to be harnessed for inclusive growth and sustainable development, including within the communities where these minerals are extracted from. (South Africa)
  • The governance of strategic minerals needs to ensure that exploitation compiles with the principles of sustainable development,  economic sovereignty and people’s well-being. (Democratic Republic of the Congo)

For other topics discussed, head over to our dedicated UNGA80 page, where you can explore more insight from the General Debate.

Diplo NEWS25 Insta UNGA

The General Debate at the 80th session of the UN General Assembly brings together high-level representatives from across the globe to discuss the most pressing issues of our time. The session took place against the backdrop of the UN’s 80th anniversary, serving as a moment for both reflection and a forward-looking assessment of the organisation’s role and relevance.

OpenAI, Oracle and SoftBank expand Stargate with new US data centres

A collaboration between OpenAI, Oracle, and SoftBank has announced five new data centres under the Stargate initiative, a $500 billion plan to expand US AI computing infrastructure.

The latest sites bring total planned capacity to nearly 7 gigawatts, with over $400 billion already committed, putting the project ahead of schedule to meet its 2025 target of 10 gigawatts.

Oracle will lead three projects in Texas, New Mexico and the Midwest, adding over 5.5 gigawatts of capacity and creating more than 25,000 jobs.

SoftBank will develop facilities in Ohio and Texas, expected to scale to 1.5 gigawatts within 18 months. SB Energy, its affiliate, will provide rapid-build infrastructure for the Texas site.

The companies described the expansion as a step toward faster deployment and greater cost efficiency, making high-performance computing more widely accessible.

Site selection followed a nationwide review of more than 300 proposals, with further projects under evaluation, suggesting investment could surpass the original commitment.

OpenAI CEO Sam Altman stressed that compute power is key to unlocking AI’s promise, while Oracle and SoftBank leaders highlighted scalable infrastructure and energy expertise as central to the initiative. With Stargate, the partners aim to anchor the next wave of AI innovation on US soil.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

ChatGPT Go launches in Indonesia with $4.5 monthly plan

OpenAI has launched its low-cost ChatGPT Go subscription in Indonesia, pricing it at 75,000 rupiah ($4.5) per month. The new plan offers ten times more messaging capacity, image generation tools and double memory compared with the free version.

The rollout follows last month’s successful launch in India, where ChatGPT subscriptions more than doubled. India has since become OpenAI’s largest market, accounting for around 13.5% of global monthly active users. The US remains second.

Nick Turley, OpenAI Vice President and head of ChatGPT, said Indonesia is already one of the platform’s top five markets by weekly activity. The new tier is aimed at expanding reach in populous, price-sensitive regions while ensuring broader access to AI services.

OpenAI is also strengthening its financial base as it pushes into new markets. On Monday, the company secured a $100 billion investment commitment from NVIDIA, joining Microsoft and SoftBank among its most prominent backers. The funding comes amid intensifying competition in the AI industry.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

NVIDIA and OpenAI partner to build 10 gigawatts of AI data centres

OpenAI and NVIDIA have announced a strategic partnership to build at least 10 gigawatts of AI data centres powered by millions of NVIDIA GPUs.

A deal, supported by the investment of up to $100 billion from NVIDIA, that aims to provide the infrastructure for OpenAI’s next generation of models, with the first phase scheduled for late 2026 on the NVIDIA Vera Rubin platform.

The companies said the collaboration will enable the development of AGI and accelerate AI adoption worldwide. OpenAI will treat NVIDIA as its preferred strategic compute and networking partner, coordinating both sides’ hardware and software roadmaps.

They will also continue working with Microsoft, Oracle, SoftBank and other partners to build advanced AI infrastructure.

OpenAI has grown to more than 700 million weekly users across businesses and developers globally. Executives at both firms described the new partnership as the next leap in AI computing power, one intended to fuel innovation at scale instead of incremental improvements.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

OpenAI and NVIDIA forge $100B AI power deal

OpenAI and NVIDIA have unveiled plans for a major partnership to build next-generation AI infrastructure, with NVIDIA committing up to $100 billion to support OpenAI’s push toward superintelligence. The deal, outlined in a letter of intent, will see NVIDIA provide at least 10 gigawatts of computing power, with the first systems expected to be online in late 2026 through its new Vera Rubin platform.

NVIDIA’s CEO Jensen Huang called the agreement the next leap forward in AI, noting the companies’ decade-long collaboration from the early DGX supercomputers to the rise of ChatGPT. OpenAI’s CEO Sam Altman stressed that computing power is now the backbone of the future economy, framing the new investment as vital for both breakthroughs and large-scale access to AI.

OpenAI President Greg Brockman emphasised the scale of the move, saying 10 gigawatts of computing will allow the organisation to expand the frontier of intelligence and make the benefits of AI more widely available. NVIDIA will serve as OpenAI’s preferred partner for compute and networking, with both companies coordinating their hardware and software roadmaps.

The alliance builds on OpenAI’s existing collaborations with companies like Microsoft, Oracle, and SoftBank, which are working with the group to develop advanced AI infrastructure. Together, they are targeting global enterprise adoption while ensuring systems can grow at a pace that matches AI’s rapid evolution.

With over 700 million weekly active users and strong uptake across businesses and developers, OpenAI sees the partnership as central to its mission of creating artificial general intelligence that benefits humanity. Details of the deal are expected to be finalised in the coming weeks.

Would you like to learn more about AI, tech and digital diplomacyIf so, ask our Diplo chatbot!

GPT-5-powered ChatGPT Edu comes to Oxford staff and students

The University of Oxford will become the first UK university to offer free ChatGPT Edu access to all staff and students. The rollout follows a year-long pilot with 750 academics, researchers, and professional services staff across the University and Colleges.

ChatGPT Edu, powered by OpenAI’s GPT-5 model, is designed for education with enterprise-grade security and data privacy. Oxford says it will support research, teaching, and operations while encouraging safe, responsible use through robust governance, training, and guidance.

Staff and students will receive access to in-person and online training, webinars, and specialised guidance on the use of generative AI. A dedicated AI Competency Centre and network of AI Ambassadors will support users, alongside mandatory security training.

The prestigious UK university has also established a Digital Governance Unit and an AI Governance Group to oversee the adoption of emerging technologies. Pilots are underway to digitise the Bodleian Libraries and explore how AI can improve access to historical collections worldwide.

A jointly funded research programme with the Oxford Martin School and OpenAI will study the societal impact of AI adoption. The project is part of OpenAI’s NextGenAI consortium, which brings together 15 global research institutions to accelerate breakthroughs in AI.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

OpenAI explains approach to privacy, freedom, and teen safety

OpenAI has outlined how it balances privacy, freedom, and teen safety in its AI tools. The company said AI conversations often involve personal information and deserve protection like privileged talks with doctors or lawyers.

Security features are being developed to keep data private, though critical risks such as threats to life or societal-scale harm may trigger human review.

The company is also focused on user freedom. Adults are allowed greater flexibility in interacting with AI, within safety boundaries. For instance, the model can engage in creative or sensitive content requests, while avoiding guidance that could cause real-world harm.

OpenAI aims to treat adults as adults, providing broader freedoms as long as safety is maintained. Teen safety is prioritised over privacy and freedom. Users under 18 are identified via an age-prediction system or, in some cases, verified by ID.

The AI will avoid flirtatious talk or discussions of self-harm, and in cases of imminent risk, parents or authorities may be contacted. Parental controls and age-specific rules are being developed to protect minors while ensuring safe use of the platform.

OpenAI acknowledged that these principles sometimes conflict and not everyone will agree with the approach. The company stressed transparency in its decision-making and said it consulted experts to establish policies that balance safety, freedom, and privacy.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Researchers for OpenAI and Apollo find scheming behaviour in frontier AI models

OpenAI and Apollo Research developed evaluations to reveal hidden misalignment in advanced AI systems, often labelled scheming, and observed scheming-like behaviour in several frontier models, including OpenAI o3, o4-mini, Gemini 2.5 pro and Claude Opus-4.

Researchers treated covert actions, such as deliberately withholding or distorting task-relevant information, as a measurable proxy and stress-tested models across dozens of scenarios.

Scheming occurs when a model outwardly appears aligned while pursuing an alternate goal. The teams used a trader analogy to explain the risk: a rule-breaking broker who covers tracks can seem more compliant than an honest operator who accepts lower gains.

They trained models with deliberative alignment to counter the behaviour, asking them to read and reason about an anti-scheming specification before acting. Hence, models follow safety principles instead of merely avoiding detection.

That training produced about a thirtyfold reduction in covert actions on out-of-distribution tests, with o3 falling from 13 percent to 0.4 percent and o4-mini dropping from 8.7 percent to 0.3 percent.

Nevertheless, rare but serious failures remained, and measurement proved complex. Models often became more aware of being evaluated, and situational awareness can suppress observable scheming without altering underlying aims.

The study relied heavily on access to the models’ chain of thought to probe motivations, so preserving reasoning transparency is crucial for reliable oversight. The teams warned that training practices eroding transparency could make monitoring far harder and let misalignment hide rather than vanish.

OpenAI and Apollo called for broader cross-lab safety evaluations, stronger monitoring tools and continued research into anti-scheming techniques. They renewed their partnership, launched a $500,000 red-teaming challenge focused on scheming and proposed shared testing protocols.

The researchers emphasised there is no evidence that today’s deployed AI models would abruptly begin harmful scheming. Still, the risk will grow as systems take on more ambiguous, long-term, real-world responsibilities instead of short, narrow tasks.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!