AI’s Dominance in 2025 Tech Developments
2025 was a year defined by rapid advances in artificial intelligence. Generative AI continued to mature from a buzzworthy novelty into a mainstream utility. In August, OpenAI launched its much-anticipated GPT-5 model, which brought upgrades to the chatbot, such as handling very lengthy prompts (up to 400K tokens) and excelling at tasks like coding and finance, prompting CEO Sam Altman to claim it felt like consulting a “PhD-level expert” on any topic. However, artificial general intelligence remained elusive. Reports noted that despite improvements in reducing hallucinations and enhancing problem-solving, GPT-5 did not meet some publicly anticipated performance expectations. Other tech giants raced to keep up: Meta, for instance, created a new “Superintelligence” research division, but later underwent an institutional reorganization and implemented a hiring freeze following significant investment in the effort.
Tech companies and governments alike poured unprecedented resources into AI R&D. In the United States, industry leaders like Google, Meta, Amazon, and Microsoft collectively planned to spend around $400 billion in 2025 on AI-related data centers and infrastructure. OpenAI itself restructured as a public benefit corporation, lifting its nonprofit constraints in order to attract more capital and push the frontier of AI without fundraising limits. The result was an AI gold rush: OpenAI’s valuation hit $500 billion, and top AI researchers were offered signing bonuses up to $100 million.
AI technology also became more embedded in everyday life. Innovations that once lived in research labs found real-world adoption in 2025. In healthcare, for example, hundreds of AI-powered medical devices have gained regulatory approval in recent years, compared to just a handful a decade ago. On city streets, self-driving taxis moved beyond pilot programs: one leading robotaxi service, Waymo, logged over 150,000 autonomous rides per week in the U.S. Across industries, an overwhelming majority of businesses reported integrating AI into their operations, up sharply from the previous year. At the same time, advanced AI became more accessible to organizations large and small. Thanks to the open-source movement, the cost of deploying a GPT-3.5-level AI system fell significantly. Open-source “open-weight” models rapidly closed the performance gap with proprietary models, shrinking the difference from 8% to under 2% on key benchmarks in the span of a year. This development reduced some of the model access barriers for smaller companies and research labs, which no longer needed to invest billions in proprietary model development. At the same time, however, open models did not eliminate all barriers to entry. Meaningful use still depends on access to substantial compute resources, specialized technical expertise, and supporting infrastructure, which can impose significant costs and constrain adoption depending on organizational needs. As a result, openness lowered some barriers while leaving others firmly in place.
Together, these trends lowered barriers and brought AI into the fabric of daily life. Not all was positive, of course; the ease of access to powerful AI also raised concerns (like making deepfakes and misinformation easier, as we’ll discuss). By the end of 2025, AI became increasingly integrated into daily life and digital infrastructure—a shift that carries profound legal, ethical, and governance implications explored in the sections that follow.
Tech Policy and Law in 2025: Toward AI Governance
As technology (and especially AI) raced ahead in 2025, governments and regulators around the world struggled to set new rules of the road. The European Union led the pack in crafting comprehensive tech legislation, most notably with its landmark AI Act. By early 2025, the AI Act was already reshaping the industry in Europe. It immediately banned certain “unacceptable” AI systems, as in those deemed to pose intolerable risks to fundamental rights or safety. For example, practices like social scoring by governments, predictive policing algorithms, or real-time biometric surveillance in public spaces became prohibited under EU law. Additionally, by February 2025 companies had to cease any such prohibited AI uses or face enforcement action in the EU. The Act also imposed new duties on AI providers: by August 2025, general-purpose AI (the kind behind chatbots and image generators) had to meet transparency requirements, including disclosing when users are interacting with AI and documenting training data for such models. These early deadlines in 2025 were just the beginning. The most stringent rules, the ones focused on “high-risk” AI systems that affect human lives in areas like healthcare, employment, education, or transportation, will go into effect in August 2026. By then, any AI used in those high-impact domains must undergo rigorous conformity assessments, be registered with EU authorities, and include human oversight mechanisms. The EU AI Act is a bold experiment that the rest of the world is watching closely.
In the United States, 2025 brought significant shifts in tech policy. In October 2023, the White House had issued an expansive Executive Order aimed at promoting “safe, secure, and trustworthy development and use of AI” by leaning on standards from the National Institute of Standards and Technology (NIST). The order also mandated federal agencies to conduct AI safety tests and audits. However, in January 2025, the White House rescinded that executive order. A new directive instructed agencies to appoint Chief AI Officers and actively remove “bureaucratic barriers” that might slow AI deployment. Instead of imposing new regulations, federal guidelines focused on voluntary standards (like the NIST AI Risk Management Framework) and on removing barriers to innovation. Under this new paradigm, baseline risk management for high-impact AI is encouraged and relies on agencies and companies to implement. On the State level, from California to Colorado, several U.S. states and even some cities proposed and passed their own AI bills in 2025.
Internationally, there was also an upswing in multilateral engagement on tech policy. Building on UNESCO’s Recommendation on AI Ethics (a global set of AI ethics principles endorsed by 194 countries in 2021), 2025 saw the launch of a Global AI Ethics Forum and growing calls at the United Nations for some form of global coordination mechanism. The idea of a global “AI watchdog” (analogous to the International Atomic Energy Agency) was floated in UN discussions over the past couple of years. One key example of international cooperation was the series of AI governance summits initiated in late 2023. After the UK’s Bletchley Park AI Safety Summit and a follow-up AI summit in Seoul, the baton passed to France in 2025. Paris hosted the “AI Action Summit,” bringing together dozens of governments, tech companies, and civil society groups to discuss the global future of AI. The Paris summit culminated in a joint pledge.
AI Ethics in 2025: Key Debates and Milestones
With AI permeating more aspects of society, ethical questions about technology moved front and center in 2025. A core theme was the human and social impact of AI, not just in abstract terms, but in very tangible ways affecting people’s livelihoods. One striking development was the growing visibility of the humans behind AI systems.
Early in the year, a group of Kenyan workers who perform data labeling for major AI systems made headlines by forming the Data Labelers Association to demand better conditions. These are the people who spend long hours tagging images, transcribing audio, and filtering toxic content to train and maintain AI models. The new Kenyan union quickly amassed hundreds of members, decrying the “systemic injustices” they say they faced from tech companies outsourcing this labor. They revealed that many were earning merely cents for complex tasks and they allege some had wages unfairly withheld. This was a reminder that ethical AI is not just about algorithms being fair or unbiased but it’s also about the ethics of how we produce AI and treat the people involved in that process. In response to mounting backlash over data annotation practices and other ethical issues, many companies have increasingly turned to synthetic data as an alternative, promoting it as a way to mitigate privacy and ethical concerns. However, as research shows, this shift does not eliminate ethical risks altogether: synthetic data can still encode bias from source data, raise unresolved questions about privacy leakage, and obscure human oversight in data production pipelines. Together, these developments stress that ethical AI is not solely a matter of fair or unbiased algorithms, but also of how AI systems are produced, whose labor is rendered visible or invisible, and which risks are displaced rather than resolved.
Concerns over algorithmic bias, fairness, and accountability remained central in the ethics discourse as well. Studies showed, for example, that facial recognition and other biometric systems remain less accurate for people with darker skin, and that large language models can produce biased content, reflecting the data they were trained on.
Responding to these concerns, the field of AI ethics has matured to actively work on interventions: new techniques for bias auditing and mitigation were proposed, and interdisciplinary collaborations (between computer scientists, social scientists, and legal scholars) expanded. Notably, in 2025 we saw efforts to operationalize ethical principles. Instead of just saying “AI should be fair or transparent,” researchers are trying to build measurable benchmarks and tools to evaluate those qualities in AI systems. A number of benchmarks, such as the FACTS benchmark (a tool for standardized testing of AI models), emerged to assess AI model behavior on dimensions like factual accuracy and safety.
A Stanford report noted that while AI-related incidents (instances of AI failures causing harm) are increasing, formal responsible AI evaluations are still rare among major tech firms. In other words, there’s a gap between recognizing risks and actually acting on them. Additionally, there’s still no consensus on what a proper fairness audit entails, and many of the metrics used are either too narrow or too easy to manipulate.
Public awareness of technology capabilities and tech ethics certainly increased in 2025, helped by open discussion and some eye-catching illustrations of ethical dilemmas. One such illustration came from the world of media. A UK television network aired a special program with a twist: at the end of the show, the female presenter revealed that she was not a real person at all, but an AI-generated avatar. The episode, while partly a stunt, drove home just how far AI-generated videos had come. The year also saw increased concern over harmful deepfake content involving minors, prompting renewed calls for stronger safeguards by platforms and regulators. These cases sharpened debates already underway in many jurisdictions about how to regulate AI-manipulated media and protect vulnerable populations.
Finally, a significant undercurrent in 2025’s tech ethics dialogue was a philosophical one: how to balance innovation and caution. On one side, AI companies were pushing exciting frontiers (like the quest for AGI). On the other side, many ethicists, researchers, and even some tech leaders urged society not to get carried away by hype and to focus on real, present-day impacts. A report by the AI Now Institute in mid-2025 argued that much of what’s wrong with the AI ecosystem boils down to concentration of power in a few big tech companies, and it warned against viewing AI progress as inevitable or inherently good. The authors advocated shifting the conversation: instead of obsessing over when superintelligent AI might emerge, ask who controls the AI we have today and who benefits from it. Indeed, some voices in 2025 called for slowing down certain AI developments (echoing the 2023 open letter that had called for a pause on very large AI experiments).
Applied Ethics Research at the Notre Dame–IBM Technology Ethics Lab
In 2025, the Notre Dame–IBM Technology Ethics Lab advanced a portfolio of applied research projects that moved beyond abstract principles to examine how ethical challenges arise in concrete human-AI interactions across domains such as peace-building, healthcare, accessibility, emergency response, and social memory. Several projects funded under the Lab’s 2025 Call for Proposals focused explicitly on the ethics of human-AI collaboration in real-world settings. For example, Building AI Text Classifiers with Peacebuilders examined how involving practitioners directly in the design of AI text classifiers can both improve technical outcomes and address ethical concerns related to legitimacy, power asymmetries, and contextual relevance in conflict analysis and resolution. Similarly, Image Descriptions Are Less Reliable Than They Appear investigated the ethical implications of AI-powered accessibility tools for blind and low-vision users, highlighting risks to user agency when AI errors are difficult or impossible to independently verify.
Other 2025 projects addressed emerging ethical questions at the boundaries of identity, care, and autonomy. Digital Afterlives? AI, Memory, and Mourning explored the rise of “ghostbots” (AI systems trained on personal data from deceased individuals) and interrogated the ethical, religious, and social implications of simulating human presence after death. Complementing this work, Digital Companionship in the Digital Age examined AI companion applications and asked whether simulated relationships with “artificial others” raise concerns about manipulation, dependency, or moral responsibility on the part of developers. In a more policy-oriented vein, Enhancing Human–AI Collaboration and Policy in Emergency Response analyzed the ethical deployment of AI-enabled drones, focusing on the tension between automation and human oversight and seeking to inform future regulatory guidance, including Federal Aviation Administration policy.
Beyond the CFP-funded projects, the Lab’s IBM Collaborative Projects further extended its ethics agenda into governance, evaluation, and accountability for large language models. Projects such as Governance, Auditing, and Risk Assessment for Large Language Models and Fairness and Equity for Large Language Models addressed the practical challenges organizations face when operationalizing ethical AI principles, including the need for interpretable metrics, stakeholder-accessible tools, and scalable governance mechanisms.
Taken together, the Lab’s 2025 research portfolio reflects a deliberate shift toward technology ethics as a lived practice, embedded in design decisions, institutional processes, and social contexts, rather than as a purely theoretical or compliance-driven exercise.
Looking Ahead: 2026 and Beyond in Tech Ethics
As we turn the page to 2026, the world finds itself at a pivotal moment in the story of technology and society. The past year’s rapid developments have brought optimism, skepticism, and anxiety, and many of those threads will continue into the coming year. One immediate focal point will be the implementation of the EU AI Act. By August 2026, any AI system deemed “high-risk” in the EU (medical AI, HR hiring algorithms, loan approval systems) must fully comply with the Act’s requirements. This represents an uncharted new phase of implementation that will reveal how the EU’s comprehensive regulatory framework operates in practice. Companies around the world that want to offer AI products in Europe are currently working on meeting these standards. How this plays out could influence other countries: if the EU’s approach shows success (preventing harms without unduly stifling innovation), it might become a model for others. If it proves too onerous or ineffective, we might see calls to rethink regulatory strategies.
In 2025, the US saw different and often conflicting strategies aimed at governing AI development and deployment. For example, several U.S. states advanced their own AI regulatory efforts, while the White House eventually issued an executive order seeking to establish a national AI policy framework. This order and framework pushed back against state AI laws deemed “onerous” and conditioned certain federal funding on state regulatory choices. The order directs federal agencies to identify and, where appropriate, legally challenge state AI regulations and promotes a “minimally burdensome national standard.” In 2026, regulatory tensions and challenges like these will likely continue as more AI-mediated technologies are deployed and more AI-related companies and startups are established.
Another area to watch is the international governance of AI. With the flurry of summits and declarations in 2023–25, the logical next step would be concrete cooperative actions. There’s talk of establishing an international panel or agency for AI (similar to the IPCC for climate or the IAEA for nuclear tech).
From the tech industry side, the large investments in AI must start showing returns soon: enterprises that spent significant capital on AI capability will want to see new revenue by 2026. If those returns aren’t obvious, we might see a pullback or more pressure on AI companies to focus on practical, narrow applications. On the other hand, the competition for AI dominance isn’t slowing down at the top. Researchers anticipate continued development of more advanced foundation models from many organizations. Industry and academic researchers are also exploring making models smaller, more domain-specific, and more efficient. Two big questions remain: Will we see any breakthroughs toward AGI? And how will the introduction of agentic AI impact markets and the future of work?
Ethically, the quest for “trustworthy AI” is not new, but it is entering a more consequential phase. Long before 2025, AI systems were already embedded, often invisibly, in domains such as judicial risk assessment, foster care, prescription review, insurance, and medical image processing. What distinguishes 2026 is not the arrival of AI in critical systems, but its growing visibility, scale, and diffusion across entire decision-making pipelines. As these systems move into public awareness, society will more clearly confront the cumulative effects of past design choices. Will companies be held accountable for harms that emerge at scale? Will regulators intervene more assertively? Or will industry self-regulation mature quickly enough to address critical issues?
One encouraging sign to watch: the continued empowerment of people in shaping tech. Whether in classrooms or courts, at dinner tables or community meetings, expect more discussions and questions of AI’s role in our lives. The more people understand AI’s strengths and limitations, the better the collective decisions we’ll make about it.
In conclusion, 2025 was a year of remarkable innovation paired with a reckoning and recognition that technology and ethics must go hand in hand. As we head into 2026, that lesson will only become more salient. The coming year promises new breakthroughs as AI systems inch closer to human-like abilities. But it also promises new challenges in helping those advances benefit humanity broadly and align with our values. The world is at a juncture where the question is not whether AI and other tech will transform society (they will), but how we manage that transformation. Will we have the wisdom and cooperation to harness technology for good while restraining its harms? The groundwork laid in 2025 – the laws enacted, the ethical debates surfaced, the public awareness raised – gives hope for a more balanced approach. Yet, the real test is about to begin. In 2026 and beyond, it’s up to all of us to ensure that technology serves humanity, and not the other way around.
Perla Khattar is a J.S.D. candidate at the University of Notre Dame Law School and a Research Fellow at the Notre Dame-IBM Tech Ethics Lab. In her research, she analyzes emerging technologies such as synthetic data, neurodata, quantum computing, and AI through a privacy lens, focusing on how law and policy can anticipate technological disruption while safeguarding individual rights.