ARCHIVED REPORTS
Overview
Planet
People
Security
Sustainability
20 Notable Reports
Directory of Reports
Editor’s Remarks
EXTRA Director of Research, Michael Marien
mdmarien@outlook.com
For better and worse, Artificial Intelligence or AI is already widespread, and still evolving, perhaps to AGI and superintelligence in the next few years. This overview seeks to identify many of the headlines and bottom lines of recent reports and articles, as well as three books—all published in 2025, with two exceptions. It is divided into three major parts:
I) The AI Race between the US and China, and a handful of massively-spending US technology organizations;
II) Impacts of AI: both current and expected; and
III) Creating Guardrails for this emerging and influential technology.
Two recent articles provide essential background to what’s happening. AI Frenzy Escalates (1) describes Meta, Microsoft, Amazon, and Google planning to spend $320 billion on infrastructure in 2025, mostly on data centers. Critics argue that there is no guarantee that this risk will live up to its potential, but many executives believe that “the bigger risk is not spending enough to keep up with rivals.” Zuckerberg Pushes Superintelligent AI (2) reports that Meta has revamped its AI strategy to include a new team dedicated to superintelligent AI that will improve “nearly every aspect of what we do.”
This investment is given a political push by Trump Intends to Unleash AI to Spur Boom (3), reporting that the US president has signed three executive orders and outlined an “AI Action Plan to “remove red tape and onerous regulation.” America started the AI race, Trump said, and “America is going to win it.” Arguably, the EU could stop Trump (4).
The investment in AI, requiring data centers, chip factories, and power supply, is pumping up the US stock market. “Companies will spend $375 billion globally in 2025 on AI infrastructure, projected to rise to $500 billion next year.” The big tech companies “are the largest financiers of the frenzy, but private equity firms have been pouring in capital, too. A major asset management firm estimates that “AI infrastructure will sop up $7 trillion over the next ten years.” (5) “If there were no data centers to build, dollars would flow into other types of investment (6).
A survey of the visions of AI firms, ”from the very plausible to the fantastical,” notes that tech companies “are taking a leap of faith,” fueled by FOMO, or “Fear of Missing Out, which does not come cheap (7).
In Empire of AI, a former investigative journalist with the Wall Street Journal describes a new age of empire where a small handful of globally scaled companies are in the field, headed by Chat GPT and OpenAI. This massively disruptive sector requires a huge number of resources to create large-language models, “arguably the most fateful tech arms race in history” (8).
In “The Economics of Superintelligence,” a July Cover Feature of The Economist (9) argues that “Many fear a hellscape, in which AI-enabled terrorists build bioweapons that kill billions, or a misaligned AI that slips its leash and outwits humanity.” But this crowds out thinking about “the immediate, probable, predictable—and equally astonishing—effects of a non-apocalyptic AI. Hence, the possibility of an explosion of economic growth, with wild swings between stocks, “as it becomes clear which companies were winning and losing winner-take-all contests.”
But these two poles of “hellscape” vs. a new economic era of global abundance are countered by recent doubts about AGI success, as well as the negative impacts of AI already visible and possible in the near future.
Gary Marcus, a founder of two AI companies and author of six AI books, argued in September that Chat GPT-5 is nowhere near the revolution many had expected, and that “the chances of AGI’s arrival by 2027 now seem remote” (10). Servaas Storm asserted in October that the US has reached “Peak Gen AI” for current large language models, and that further scaling of chips and data centers will not deliver AGI (11).
Some negative impacts are already visible. A “flood of fake photos and images” has amplified social and partisan divisions and bolstered antigovernment sentiment (12). OpenAI’s new “Sora” smartphone app enables the creation of videos entirely from AI, making disinformation easier and endless (13). “Be Wary of Chatbots Offering Guidance” (NY Times, 30 Sept, 2025, D6) warns of “falling prey to AI’s flattery” and AI companions that lead to social deskilling.
Cyberattacks are escalating in speed, volume, and sophistication. GenAI chatbots serving as a “force multiplier” for the global hacker toolbox (14). A report from RAND Europe warns of uncontrollable AI incidents and calls for mandatory security audits and independent oversight (15). Even scarier, an 80-page April report from Forethought warns of AI-enabled coups to seize power, reinforced by a 40-page July report from RAND on an AGI Coup as one of several scenarios (16).
An excellent overview is provided by an Elon University survey of 301 experts on “Being Human in 2035” (17), where positive changes are seen in the capacity to learn and problem-solving, with negative changes in social and emotional intelligence, capacity to think deeply, mental well-being, and sense of identity and purpose. Fundamental change in human capacities was expected by 23%, considerable change by 38%, moderate change by 31%, and little or no change by 8%. Overall, 16% saw AI as mostly for the better for most people worldwide, 50% saw equal changes for better and for worse, 23% saw changes mostly for the worse, 5% saw little or no change, and 5% don’t know.
Finally, if all of this seems too much to handle, a provocative new book by Eliezer Yudkowsky and Nate Soares, leaders of the Machine Intelligence Research Institute in Berkeley, argues that “If Anyone Builds It Everyone Dies: Why Superhuman AI Would Kill Us All”, and that the stakes are existential (18).
In August, the UN General Assembly established an Independent Scientific Panel on AI, and an annual Global Dialogue on AI Governance, recognizing the need to regulate AI “before it becomes a threat to world stability and peace.” However, permanent members of the outdated Security Council can block initiatives to rein in the technology. Such members include China and the US, the leading competitors in AI development. The Director of the White House Office of Science and Technology, Michael Kratsios, said on Sept 24 that “We totally reject all efforts by international bodies to assert centralized control in global governance of AI” (19).
Three other UN bodies have advised on developing guardrails:
Other organizations have also urged various guardrails:
In sum, there are plenty of thoughtful organizations to advise the UN’s Independent Scientific Panel on AI, and to participate in the annual Global Dialogue on AI Governance.
A May 2025 article in Scientific American asked, “Could AI Really Kill Off Humans?” (31). It concludes that “no scenario can be described where AI is conclusively an extinction threat to humanity,” noting that it is very hard to kill all of us, through nuclear war, biological pathogens, or climate change. But it is potentially easy to kill off many billions of us through any number of AI-related possible catastrophes. Effective global governance of AI is obviously needed–the sooner the better, but probably too little, too late. This forecast will hopefully be wrong.
On a somewhat more positive note, the first “Time 100AI” on “The 100 Most Influential People in Artificial Intelligence” provides brief descriptions of “Leaders, Innovations, Shapers, and Thinkers”, both supporters and critics (32).
Human Development Report 2025
United Nations Development Programme
March 2025, 324p. Explores the profound, dual-edged impact of artificial intelligence (AI) on human development, noting breakthroughs in creativity and productivity alongside risks of bias, inequality, and ethical dilemmas. Finds that “AI is increasingly enabling cross-border collaboration in research and innovation, fostering new networks of knowledge production across regions” but warns of existential risks, recommending balanced governance to promote inclusion, equity, and resilient systems. Key recommendations: Foster inclusive, trustworthy AI development; address IP challenges; implement equity-focused regulatory frameworks.
Mapping AI Risk Mitigations: Evidence Scan and Draft Mitigation Taxonomy
MIT AI Risk Initiative
March 2025, 24p. Synthesizes 831 risk mitigations from 13 frameworks, categorizing them into “Governance & Oversight, Technical & Security, Operational Process, and Transparency & Accountability Controls.” Stresses operational safeguards, continuous monitoring, and that “AI risk management is an emerging concept,” serving as a foundational resource for global decision-makers. Key recommendations: Prioritize ongoing monitoring and robust risk processes; encourage community and stakeholder feedback on actionable strategies.
Superintelligence Strategy: Expert Version
Center for AI Safety
March 2025, 41p. Frames superintelligent AI as a strategic and security challenge. Details the Mutual Assured AI Malfunction (MAIM) deterrence framework, highlights chip vulnerability and supply chain risks, and stresses the need for legal, multipolar governance frameworks. “Outcomes hinge on what we do next” underlines the urgent importance of coordinated deterrence. Key recommendations: Strengthen supply chains, legal frameworks, and international deterrence to prevent the uncontrolled escalation of superintelligence.
The Artificial General Intelligence Race and International Security
Sarah Kreps et al., RAND Corp & Perry World House
Sept 2025, 72p. Assesses AGI’s impact on international stability, focusing on U.S.–China competition and the transition phase before AGI’s maturity. Notes the inadequacy of traditional arms control for AGI’s dual-use nature and proposes innovative international governance approaches, including an “AI cartel”. Warns that “risks arise not only from AGI’s eventual power, but also critically from the ambiguous and volatile period preceding its arrival.” Key recommendations: Develop tailored governance for dual-use technology; maintain strategic communication and flexibility during AGI’s formative phase.
How Artificial General Intelligence Could Affect the Rise and Fall of Nations: Visions for Potential AGI Futures
Barry Pavel et al., RAND Corp
July 2025, 40p. Explores scenarios for AGI’s influence on global power, including the “New Renaissance” through cooperative innovation versus “Governance Failure” or an “AGI Coup” driven by misaligned, centralized superintelligence. Identifies existential risks, including economic, security, and authoritarian shifts. Stresses the importance of public-private partnerships and robust alliance structures for alignment and safety. Key recommendations Include Promoting balanced oversight, resilient alliances, and proactive safety and ethics protocols.
Strengthening Emergency Preparedness and Response for AI Loss of Control Incidents
Elika Somani et al., RAND Europe
Aug 2025, 61p. Calls for multi-layered strategies to prevent uncontrollable AI incidents, recommending mandatory reporting, security audits, independent oversight, and a safety-first development culture. Notes preparedness must precede deployment, addressing open-source risks and emphasizing information sharing. Key recommendations: Institute mandatory reporting, routine audits, and joint risk management across stakeholder groups.
The Singapore Consensus on Global AI Safety Research Priorities
Singapore Conference & Infocomm Media Development Authority
May 2025, 33p. Synthesizes expert agreement on defense-in-depth measures for safe GPAI systems, spanning risk assessment, alignment, robustness, and real-time control and intervention mechanisms. Identifies cooperation in risk thresholds, incident response, and dynamic benchmarking as ways to reduce collective harm and guide global AI safety R&D. Key recommendations: Establish shared risk standards, rapid incident response, and adaptive institutional structures to manage cascading AI impacts.
Artificial Intelligence and Peacebuilding: Opportunities and Challenges
International Panel on the Information Environment
Sept 2025, 60p. Analyzes AI’s dual role in peacebuilding, enhancing conflict analysis and citizen engagement while highlighting bias, misinformation, and ethical risks. Advocates rights-based, conflict-sensitive AI design, local participation, and ongoing risk assessment, emphasizing human oversight and contextual adaptation. Key recommendations: Design inclusive and transparent AI solutions with local input; foster international cooperation and cross-sectoral risk management for fragile states.
Multi-Stakeholder Forum on Science, Tech, and Innovation for the SDGs
UN Economic and Social Council
31 May 2024, 19p Summary. Solutions and innovations to support progress across the SDGs with a focus on Goals 1, 2, 13, 16, and 17, with >300 scientists submitting briefs and 99 passing peer review. Emphasis on AI driving precision farms to increase yields up to 70% by 2050 and AI potential in health care—but AI data centers consume 1% of global electricity and use large amounts of freshwater.
Summit of the Future Outcome Documents: Pact for the Future, Global Digital Compact, and Declaration on Future Generations
United Nations
Sept 2024, 66p. Final version of the Pact, listing 56 Actions on sustainable development and financing (“we will take bold, ambitious, accelerated, just and transformative actions to implement the 2030 Agenda”), international peace and security (“we will redouble our efforts to build and sustain peaceful, inclusive and just societies and address the root causes of conflicts”), science and technology, youth and future generations, and global governance. The Global Digital Compact (pp40-55) seeks to “close all digital divides” and enhance AI governance. The Declaration seeks stronger youth participation.
Crowdstrike 2025 Global Threat Report
CrowdStrike
April 2025, 53p. “Cyberattacks are escalating in speed, volume, and sophistication.” Identifies a shift in 2024 toward streamlined, scalable attacks driven by a business-like approach. “Don’t underestimate today’s enterprising adversaries,” with a “force multiplier” impact of off-the-shelf chatbots making genAI “a popular addition to the global hacker toolbox.” In 2024, China-nexus activity surged 150% across all sectors. Voice-phishing (vishing) attacks skyrocketed, with the average e-crime breakout time averaging 48 minutes. Most detections were malware-free. Access broker ads increased, and 26 new adversaries tracked by CrowdStrike raised the total to 257. North America had 53% of interactive intrusions, followed by 14% in Russia, 11% in Europe, and 7% in India. Emphasizes the need for proactive defense strategies and adaptive cyber resilience as threat actors grow more agile and commercially motivated.
Coalition for Sustainable AI
AI Action Summit
Feb 2025, 5p. Launched at the Paris AI Action Summit, the Coalition, initiated by France, UNEP, and the ITU, with >100 partners in 11 countries, unites global stakeholders to advance AI’s alignment with environmental and climate goals while addressing its environmental impact. It provides a “Platform of Engagement” to connect stakeholders and an “Initiatives Hub” to enhance collaboration, visibility, and avoid duplication. The Coalition supports the UN’s SDGs and coordinates with existing initiatives, “continuing work for an AI respectful of Planetary Boundaries.” The transformative potential of AI in tackling the climate and environmental crisis is already unfolding. Still, the environmental footprint of AI is also growing, and there is “a risk of fragmented and redundant initiatives that dilute impact.”
Being Human in 2035: How Are We Changing in the Age of AI?
Imagining the Digital Future Center (Elon Univ, Elon NC)
April 2025, 286p. A survey of 301 experts asked to predict AI impact by 2035 on 12 essential human traits and capabilities. Change is likely to be primarily positive in curiosity and capacity to learn, decision-making and problem-solving, and innovative thinking. Change is likely to be mostly negative in social and emotional intelligence, capacity to think deeply, trust in widely shared values and norms, mental well-being, sense of identity and purpose, etc. Dramatic and fundamental change in human capacities as advanced AI is broadly adapted was expected by 23% of the experts; considerable change by 38%, moderate but noticeable change by 31%, minor and barely perceptible change by 5%, and no noticeable change by 3%. Overall, 16% view AI as mostly beneficial for most people worldwide, 50% see fairly equal changes for better and worse, 23% believe changes will mostly be for the worse for most people, 6% expect little to no change overall, and 5% are unsure.
Empire of AI: Dreams and Nightmares in Sam Altman’s Open AI
Karen Hao (former Wall Street Journal writer)
Penguin, May 2025, 496p, $32. (Pub in UK by Allen Lane, as “Empire of AI: Inside the Reckless Race for Total Domination.”). An AI expert and investigative journalist describe a new and ominous age of empire, where a small handful of globally scaled companies are at the forefront, led by ChatGPT and OpenAI. The vision of success for this massively disruptive sector requires a huge number of resources to create massive large-language models, “arguably the most fateful tech arms race in history.”
Governing AI for Humanity
UN High-Level Advisory Body on AI
7 July 2024, 95p. AI governance is “crucial” to address challenges and risks, and ensure its “tremendous potential” is realized. Despite much discussion, “the patchwork of norm and institutions is still nascent and full of gaps.” On common ground and benefits, coordination and implementation gaps, a UN AI Office, etc.
Framework Convention on Global AI Challenges: Accelerating International Cooperation to Ensure Beneficial, Safe and Inclusive AI
Centre for International Governance Innovation (Waterloo, ON, Canada)
June 2024, 34p. CIGI’s Global AI Risks Initiative is concerned with loss of control of advanced AI systems and AI weaponization (misuse of AI systems to cause harm). A Framework Convention should codify the most important shared objectives for AI cooperation in addressing the most urgent issues posed by “accelerating development of AI.”
Artificial Intelligence Governance to Reinforce the 2030 Agenda
UN Economic and Social Council
29 Jan 2024, 16p. On AI’s potential to accelerate progress in poverty reduction, education, and other areas, as well as its associated risks, including job displacement and data bias. Stresses the urgent need for governance, as AI rapidly advances to AGI and potential existential risks.
Requirements for Global Governance of Artificial General Intelligence: Phase 2 of a Real-Time Delphi
The Millennium Project (Jerome Glenn, CEO)
April 2024, 46p. AGI is an advanced AI capable of autonomous learning across domains. Most experts project the emergence of AGI within 3-5 years, followed by Artificial Superintelligence (ASI). Many of the 229 respondents warn of existential risks from unregulated AGI.
Managing Extreme AI Risks Amid Rapid Progress
Science, Yoshua Bengio, G. Hinton, and 23 Others
20 May 2024, 5p. AI is progressing rapidly, as companies shift focus to generalist AI systems that act autonomously, but with risks including large-scale social harms, malicious uses, and loss of human control. “AI safety research is lagging.” Governance measures must prepare us for sudden AI breakthroughs with an automatic trigger when AI hits certain milestones.
The AI Risk Repository: A Comprehensive Meta-Review, Database, and Taxonomy of Risks
MIT Future Tech
Aug 2024, 79p. Classifies 777 risks into 7 AI risk domains and 23 sub-domains.
The Digitalist Papers: Artificial Intelligence and Democracy in America
Stanford Digital Economy Lab
24 Sept 2024, 240p. $36.72; $28.97pb from Amazon. Just as the Federalist Papers of the 18th century analyzed the great challenges of the day, these 12 essays by 19 “thought leaders” consider new challenges to democracy and participatory practices, AI’s potential to transform government operations and public service, the complex challenges of AI regulation, and the need for participatory frameworks and ethical considerations.
AI Landscapes: Exploring Future Scenarios of AI to 2030
Rachel Adams and 20 Others, Economist Impact
Feb 2024, 29p. Four AI development scenarios, ranging from global cooperation to fragmented policies.
The AI Revolution: What the New Age of Artificial Intelligence Means for Humanity
NewScientist Essential Guide No. 23
July 2024, $15. Why has AI suddenly leapt forward? Describes how the technology works, its capabilities, and “future horizons from utopia to annihilation.”
The Age of AI: And Our Human Future
Henry A. Kissinger, Eric Schmidt (former Google CEO & Chair), and Daniel Huttenlocher, Little, Brown and Company
Oct 2021, $30. Discusses the emerging human-machine partnership, the evolution of AI, the dream of AGI, global network platforms and disinformation, security and world order, conflict in the digital age, AI and the international order, managing AI, human identity and AI, and the essential need for an AI ethic.
Technology as a Force for Good: Technology Driving the Transition to a Superior Future
Force for Good, Ketan Patel
Jan 2024, 122p. Warns of the polycrisis, “a cascade of successive global disruptions diverting leaders’ attention and resources away from longer term systemic priorities.” Shows that 19 core technologies, now existing, can enable necessary transitions and advance the SDGs.
