Expert SpeakDigital Frontiers
Published on Oct 31, 2026
AI safety is increasingly entangled with corporate interests and US electoral politics, raising questions about independent oversight, evidence-based regulation and public accountability
In September 2026, a former researcher at OpenAI and Anthropic resigned from the latter with an extraordinary warning. He alleged that the two companies were racing towards self-improving superintelligence and “gambling with our lives”. He argued that future systems could become difficult to control,“could kill us all” and called for government regulation and independent auditing. Since then, the political landscape in the US has also rapidly engaged with this debate, with both Democrats and Republicans advocating different approaches to artificial intelligence (AI) regulation and deregulation.
However, calls for regulation from frontier AI companies may not always serve the wider industry’s interests. Depending on its design, regulation can entrench the market power of the very companies it seeks to constrain. New regulations impose compliance costs that established AI laboratories can absorb more readily than smaller firms and new entrants. These requirements can therefore become barriers to entry, shielding dominant companies from competition while limiting the scope for innovation across the industry.
Calls for regulation from frontier AI companies may not always serve the wider industry’s interests. Depending on its design, regulation can entrench the market power of the very companies it seeks to constrain.
This is particularly important given that the large-language-model (LLM) paradigm dominant today may not shape the next phase of AI development. Additionally, current evidence does not establish that LLM capabilities have necessarily plateaued. However, existing evaluations may not reflect the absolute capabilities of frontier models. This progress is also becoming increasingly resource-intensive and cost-sensitive. Estimates suggest that frontier AI training costs are increasing approximately 2.4-fold each year, and memory components are becoming increasingly scarce.
Thus, as the economic and energy stakes grow, debates over AI development have become increasingly entangled with broader political and moral divisions, particularly in the United States. Democrats have generally emphasised stronger regulatory safeguards, human oversight and corporate accountability, while Republicans have prioritised technological leadership, competition and fewer regulatory barriers, framing rapid AI development as essential to American economic and national security interests.
Meanwhile, US media coverage has devoted considerable attention to AI’s risks, ranging from catastrophic scenarios involving human extinction to potentially harmful or malicious behaviour by autonomous agents and the projected consequences of military applications. These concerns warrant scrutiny, but sensational framing can blur the distinction between demonstrated harms, failures observed in controlled experiments and speculative future risk scenarios.
Mixed signals From the Frontier
Leading AI laboratories have spent years promoting two apparently contradictory narratives. The first promises systems of unprecedented economic and scientific value: artificial general intelligence, superintelligence and models capable of discoveries beyond human researchers. The second warns that these same systems could become uncontrollable, urgently requiring government intervention.
These safety narratives are further complicated by OpenAI’s recent announcement that its systems had solved the Navier–Stokes equation and more than 372 other long-standing, open mathematical problems. If independently validated, this would represent a major scientific achievement at a scale and speed rarely seen in scientific research. It is nevertheless also a powerful commercial and national signal: OpenAI is telling governments, investors and competitors that its unreleased systems have entered an entirely new capability regime.
When an AI company warns that its forthcoming systems may be too powerful to develop under ordinary market rules, it simultaneously advertises the exceptional value of those systems. Safety warnings may therefore be less sincere and more commercially advantageous. This deep-seated tension has become more consequential as enormous amounts of capital depend on expectations of continued and sometimes exponential progress. For example, OpenAI submitted a confidential draft registration statement for a possible initial public offering in June 2026, although it did not commit to a timetable. Anthropic has also filed paperwork for a preliminary IPO while advocating different forms of AI oversight.
AI Enters Electoral Politics
The safety debate is intensifying ahead of the 2026 US midterm elections. President Donald Trump’s policy has explicitly favoured rapid development and the removal of regulatory barriers to maintain and reinforce American leadership in AI. Its AI policy seeks American technological dominance through accelerated innovation, infrastructure expansion and private-sector competition. The administration has argued that excessive regulation could impede American companies and strengthen China.
This creates a potential opening for Democrats to contrast Republican deregulation with a focus on public and environmental protection, safety and privacy. Concerns over job displacement, rising electricity prices, data-centre construction, misinformation and concentrated corporate power can all be organised around AI governance and misgovernance. Democratic leaders therefore portray the administration’s policies as insufficiently attentive to public risk and argue that they empower wealthy technology companies, potentially broadening economic inequality and exacerbating the cost-of-living crisis.
Yet it would be analytically unsound to describe the renewed regulatory push as a coordinated Democratic operation against the Republicans. Available evidence does not establish that Coxon’s resignation or the companies’ subsequent interventions were designed to influence the midterms. Nor is the political division completely partisan. Democrats disagree among themselves about regulation, while some conservative figures also support stronger controls. Recent evidence has also shown that, both parties broadly agree on the importance of responsible innovation.
Technology companies are also intervening across party lines. Corporate political spending has reached record levels during the 2026 election cycle, with AI emerging as one of its major sources. For example, political groups connected to OpenAI and Anthropic, or their executives, spent more than US$23 million supporting two competing Democratic candidates in a single New York congressional contest. The companies were not necessarily advancing the same regulatory position. This indicates that AI firms are not merely responding to politics; they are increasingly participating in its production.
What Politicisation Costs
Politicisation can ultimately damage AI safety research. Dramatic extinction scenarios dominate public attention and increase fear, while more measurable but immediate problems receive less scrutiny. Conversely, treating every safety proposal as an anti-innovation manoeuvre makes it difficult to establish even narrowly targeted safeguards for critical scenarios.
Politicisation can also shape which safety questions researchers are able to investigate.When funding and access to frontier models depend heavily on developers or governments pursuing strategic advantage, research agendas risk reflecting those institutions’ priorities. Questions about labour, local deployment conditions or less commercially attractive safeguards may receive less support. Independent evaluation therefore requires more than permission to test a model: it requires sustained funding and sufficient access for researchers to pursue questions that developers and political sponsors may have little incentive to ask.
Politicisation can also shape which safety questions researchers are able to investigate.
However, the answer is neither indiscriminate, blanket deregulation nor regulation constructed around viral warnings and unverified claims. Governments should require frontier developers to conduct standardised capability evaluations, disclose serious and unexpected incidents and promote genuinely independent testing. Regulatory thresholds should be tied to demonstrated, real-world capabilities, such as autonomous cyber operations, autonomous weapons and AI-assisted drug design, rather than elastic terms such as “AGI” or “superintelligence”. Most importantly, claims made by frontier laboratories must be treated as evidence to be tested, not as neutral descriptions of reality.
When AI discourse becomes a political tool, fear, scientific achievement and regulatory policy can be compressed into a single narrative serving electoral or corporate interests. This risks deepening polarisation across generations and economic groups, as people interpret technological developments through existing political loyalties and concerns about their livelihoods. Citizens may also lack access to the technical knowledge and reliable information needed to assess what new developments mean or what AI systems can actually do.
The purpose of AI safety should be to make existing and emerging systems more accountable and explainable, not to make their developers or the political parties backing them more influential.
Governments and the media therefore have a responsibility to communicate capabilities, limitations and uncertainties clearly, distinguish demonstrated harms from speculative risks, and subject corporate claims to independent scrutiny. The public, in turn, should have opportunities to question these claims and participate meaningfully in decisions about AI’s development and use. The purpose of AI safety should be to make existing and emerging systems more accountable and explainable, not to make their developers or the political parties backing them more influential.
Ishita Deshmukhis a Research Assistant at Observer Research Foundation.
The views expressed above belong to the author(s). ORF research and analyses now available on Telegram! Click here to access our curated content — blogs, longforms and interviews.
Author
Ishita Deshmukh
Ishita Deshmukh is a Research Assistant with ORF’s Centre for Security, Strategy & Technology.
Her work focuses on how artificial intelligence is reshaping national security, economic …
