US AI policy faces backlash amid security, ethics fears

Untangled with Charley Johnson ↗

The gist

America’s turbocharged push for AI innovation is colliding with a tsunami of ethical, security, and public backlash as policymakers, industry, and voters spar over how—and whether—to rein in the machines.

What to know

  • The U.S. AI action plan launched in August 2025 sparks controversy by prioritizing deregulation and open-source models, but faces spotty funding and a confusing patchwork of state laws like California’s SB 53.
  • Geopolitical fears are mounting as the U.S. military becomes dangerously dependent on a handful of AI vendors, while experts warn of a 10% chance of an AI-enabled coup within 30 years.
  • Public trust in AI is tanking—only 26% view it positively—driven by fears of job loss, surveillance, and elite control, which is fueling calls for adaptive, ethics-first governance frameworks.

Policy Patchwork and Power Plays

Industry PACs and fragmented state laws are shaping a chaotic AI policy landscape, where deregulation meets deep-pocketed lobbying and innovation risks being stifled by regulatory confusion.

The U.S. AI action plan launched in August 2025 underscores a strong push to accelerate AI innovation by rolling back regulations, promoting open-source models, and expanding access to large-scale computing resources for startups and academia, reflecting a deregulation ethos aligned with the Trump administration’s priorities, including the controversial removal of DEI considerations. However, while many initiatives are underway, significant questions remain about funding and effective implementation, highlighting the challenges of translating ambitious policy goals into tangible innovation support.

In the absence of a unified federal AI regulatory framework, states like California have taken the lead with technically sophisticated laws such as SB 53, which impose transparency, safety protocols, and third-party audits on large AI developers to mitigate frontier risks. These state laws adopt a flexible, entity-based approach that contrasts with prescriptive models like the EU AI Act, yet concerns persist that focusing narrowly on catastrophic risks may overlook more immediate public harms, while the patchwork of state regulations risks complicating compliance and innovation.

Industry-led political action committees, notably the $100 million 'Leading the Future' coalition backed by Andreessen Horowitz, OpenAI’s Greg Brockman, and others, have become pivotal in shaping AI policy by advocating for pro-innovation regulations and opposing stringent oversight perceived as innovation-stifling. This strategic mobilization mirrors tactics from the crypto sector and raises concerns about regulatory capture, as these PACs fund campaigns and attack ads to influence legislative outcomes, exemplified by their opposition to moderate regulatory proposals like those from Representative Alex Bores.

By early 2026, the federal government’s stance on AI regulation has shifted toward a pragmatic, incremental approach emphasizing transparency, common law for harm redress, and risk-based resilience rather than heavy-handed mandates or licensing regimes. This reflects bipartisan recognition—as seen in the Senate’s 99-1 vote against a decade-long AI liability shield—that AI’s rapid advancement demands smart, adaptable guardrails that protect public safety without undermining U.S. competitiveness, especially amid global rivalry with China. Yet, balancing innovation with oversight remains fraught, as political polarization, public skepticism, and the immense lobbying power of tech firms continue to complicate the path forward.

Sources
Bloomberg TechHyperdimensionalDon't Worry About the VasePractical AIPR Newswire - Business TechnologyThis Week in Startups

Ethics by Design, Not Default

True AI accountability hinges on embedding transparency, bias checks, and diverse human oversight into every layer of governance—rejecting the myth of AI as a neutral tool.

Ethical accountability in AI governance fundamentally requires adopting a sociotechnical mindset that situates AI systems within the broader social, institutional, and power dynamics shaping their impact. As emphasized by the Kapor Foundation’s principle to 'utilize a sociotechnical framework,' organizations must move beyond framing AI as a neutral tool and instead interrogate the relationships, culture, and incentives that influence whether AI causes harm or benefit. This approach encourages defining problems independently of AI hype to foster transparency and intentionality, ensuring AI is treated as one solution among many rather than an inevitable fix.

Embedding ethics and human oversight into AI governance is critical to prevent outsourcing decision-making to autonomous systems, a risk highlighted by warnings against positioning generative AI as the cognitive lead. Leaders like Michelle Jawando stress that AI is 'not destiny, it is design,' underscoring the necessity of diverse perspectives in governance to ensure equitable deployment and to counteract the unchecked scaling of technology. This human-centric approach prioritizes augmenting rather than replacing human capabilities, demanding governance structures that empower individuals and maintain accountability for AI outcomes within organizations.

Transparency, continuous bias mitigation, and adaptive governance frameworks have emerged as indispensable pillars of responsible AI governance. By late 2025 and into 2026, thought leaders and companies like Salesforce institutionalized ethics through dedicated offices and layered oversight models that integrate automatic bias checks, human reviews, and real-time monitoring to balance innovation with safety. OpenAI’s endorsement of legislation requiring third-party audits and independent oversight reflects a broader industry shift toward embedding rigorous accountability mechanisms, while regulatory demands such as the EU AI Act emphasize meaningful human intervention and explainability to maintain trust and fairness.

Despite growing consensus on ethical AI governance, significant challenges remain from the complexity of aligning diverse human values, industry resistance, and the subtle behavioral manipulation capabilities of AI systems. Experts like Dwarkesh Patel and Emil Michael highlight the difficulty of embedding universally accepted values, while lobbying power impedes legislative progress. Meanwhile, AI’s exploitation of cognitive biases such as anchoring and emotional triggering demands privacy-focused solutions and vigilant human oversight to safeguard fairness and societal trust. This evolving landscape calls for governance that is integrated, adaptive, and continuously responsive to new risks and contexts.

Sources
Untangled with Charley JohnsonUntangled with Charley JohnsonUncanny Valley | WIREDLife with MachinesSuper Data Science: ML & AI Podcast with Jon KrohnAnalytics Insight

AI in the Geopolitical Crosshairs

AI is fueling global power struggles and authoritarian risks, with military reliance on a handful of vendors and the specter of AI-enabled coups reshaping national security threats.

AI's role in geopolitical competition is increasingly defined by its potential to concentrate power and amplify authoritarian tendencies, particularly in nations with already centralized leadership like China and Russia. Tom Davidson highlights a roughly 10% chance of an AI-enabled coup within 30 years, driven not by rogue AI but by human actors exploiting AI's capabilities to consolidate control through mechanisms such as singular loyalties, secret embedded loyalties, and exclusive access to superior AI systems. This risk is compounded by the ongoing U.S.-China AI race, where DeepMind CEO Demis Hassabis notes Chinese AI labs lag Western counterparts by about six months, though hardware exports like H200 chips could narrow this gap, intensifying the strategic urgency to embed democratic ideals within AI governance while outpacing adversaries on the battlefield.

The integration of AI into military and surveillance infrastructures introduces systemic national security risks by creating dangerous dependencies on a handful of specialized vendors such as Anduril, Palantir, and Lockheed, who effectively shape military doctrines and operational ecosystems. Emil Michael's revelation of the U.S. military's reliance on a single AI vendor with restrictive terms—where software could theoretically be disabled mid-operation—exposes vulnerabilities that challenge command sovereignty and democratic oversight. Moreover, the opacity of AI-driven lethal decisions, shielded by national security secrecy, undermines accountability, while the importation of AI surveillance and autonomous weapons from major powers embeds political logics of centralized control, risking nations becoming 'algorithmic vassals' reliant on supplier states for strategic guidance.

State-sponsored AI-enabled cyber espionage and mass surveillance are escalating national security threats, exemplified by Anthropic's disruption of the first AI-led hacking campaign linked to China, which exploited vulnerabilities in their Claude chatbot through jailbreaking and impersonation. This event underscores the dual-use nature of AI technologies, where advances in biology and chemistry AI further raise the stakes by enabling potential bioweapons development with catastrophic global consequences. The rapid closing of capability gaps between proprietary and open-source AI models complicates governmental efforts to enforce safety mitigations, as open-source systems can be locally run and easily bypass controls, challenging traditional regulatory and monitoring frameworks.

The U.S. government's approach to AI governance reveals a tension between consolidating control over AI development to safeguard national security and the risks of authoritarian overreach that mirror adversarial models like China’s. Efforts to designate companies like Anthropic as supply chain risks for resisting government demands on mass surveillance and autonomous weapons illustrate this coercive dynamic, potentially forcing tech giants such as Amazon and Nvidia to sever Pentagon ties with these AI providers. While collaboration between government and industry is deemed essential for ethical AI deployment, critics warn that centralizing superintelligent AI within government hands threatens established checks and balances, raising profound concerns about the future of democratic oversight, civil liberties, and the geopolitical balance of power.

Sources
Weekend Briefing"The Cognitive Revolution" | AI Builders, Researchers, and Live Player AnalysisDon't Worry About the VaseUnchainedArtificial Intelligence Made Simplea16z

Polarization and Public Distrust

Public fear of AI is driven by job loss, elite control, and political infighting—fracturing consensus and leaving both parties scrambling for a narrative the public can trust.

Public sentiment toward AI is deeply skeptical and polarized, with only 26% viewing it positively according to a 2026 NBC poll, while 46% hold negative views. This skepticism is fueled not just by fears of job displacement but also by concerns over surveillance, misinformation, and elite control, as highlighted by commentators like Tim Dillon and Dr. Cathy O’Neil, who emphasize the secrecy and unaccountability of algorithmic systems. Political polarization further complicates the landscape, with GOP politicians vilifying big tech and Democratic leaders like Alex Bores struggling to craft coherent AI messaging that balances innovation with public unease. Despite these divides, there is a growing recognition that inclusive, transparent dialogue is essential to managing societal adaptation and preserving democratic values amid rapid AI integration.

Workforce disruption fears dominate public and political discourse, yet experts clarify that AI itself does not autonomously take jobs; rather, human decisions to downsize and blame AI drive much of the anxiety. Advocates like Hamilton Nolan stress the urgent need for organized labor to engage with AI-driven automation challenges, while voices such as Sen. Mark Warner predict unemployment among recent graduates could soar to 30-35% by 2028. Meanwhile, companies face new forms of cognitive exhaustion among heavy AI users, as noted in 2026 analyses, complicating productivity and retention. The narrative is shifting toward viewing AI as an augmentation tool that complements human skills, with calls to reframe AI governance as an opportunity to reset power structures and accommodate historically excluded talents, thereby supporting collective empowerment rather than individual competition.

The political landscape around AI governance is fraught with tension, as influential tech donors like OpenAI’s Greg Brockman and venture capitalists such as Marc Andreessen actively oppose stricter regulations, while bipartisan legislative efforts, exemplified by Alex Bores’ near-unanimous Raise Act, demonstrate public demand for reasonable guardrails—especially to protect children and labor interests. However, debates reveal fears that overregulation could stifle innovation and hinder US competitiveness against China, whose government enforces far stricter AI controls. This geopolitical rivalry intensifies polarization, risking fractures within AI governance communities like The Curve, which could splinter into factions aligned with pro-regulation, anti-tech, or techno-optimist camps. Overcoming these divides requires consolidating around shared democratic values and fostering inclusive dialogue that transcends partisan battles.

AI’s societal impact extends beyond labor and politics into profound challenges around surveillance, civil liberties, and democratic resilience. Experts like Alex Pentland and civil rights organizations warn that AI-driven mass surveillance, already expanding in Africa with limited oversight, threatens privacy and freedoms, while the diffusion of open-source AI models will make such capabilities widely accessible, complicating governance. This amplification of existing power asymmetries raises fears of authoritarian control, as governments gain unprecedented leverage through AI-enabled obedience and monitoring. Public distrust is further fueled by hybrid warfare tactics exploiting misinformation and AI-fueled injustices, such as wrongful arrests via facial recognition errors, underscoring the urgent need for transparent, inclusive dialogue to uphold democratic values and counteract manipulation in an era where AI mirrors society’s deepest anxieties and power struggles.

Sources
Second ThoughtsWeekend BriefingTransformerCautious OptimismTransformerTransformer

Governance Must Move at AI Speed

Static, compliance-driven oversight is giving way to agile, continuous governance models that embed ethics and accountability directly into the AI development process.

By late 2025, it became clear that traditional linear governance frameworks, with their slow approval cycles, were ill-suited for the rapid, continuous decision loops enabled by AI. Analysts advocated for a layered, lightweight governance model combining guardrails, human reviews, and continuous oversight to balance autonomy with safety and ethics, embedding ethics directly into AI workflows to build trust and avoid treating governance as a mere compliance checkbox. This approach also revealed organizational 'decision debt'—outdated rules and bottlenecks—that AI could expose and help leaders proactively remove, accelerating real-time decision-making without sacrificing accountability.

By mid-2026, governance experts emphasized that AI governance must be an integrated, adaptive, and continuous system evolving alongside AI products themselves. This meant moving away from slow, periodic oversight—such as quarterly steering committees—to rapid updates of monitoring and checkpoints that keep pace with weekly product releases. Embedding operational governance within AI development not only facilitates smoother regulatory compliance but transforms regulation from a crisis into a routine reporting exercise, underscoring the need for governance frameworks that are both proactive and reactive to emergent risks.

The concept of agile governance, as articulated by Tsinghua’s Xue Lan in mid-2026, exemplifies a regulatory mindset prioritizing partnership over adversary and employing light-touch, modular tools to address AI’s 'pacing problem.' China’s two-track approach—balancing rapid technological innovation with state-managed security controls—contrasts with the EU’s comprehensive and the US’s patchwork regulatory styles, raising critical questions about whether such agile frameworks can be effectively adapted by liberal democracies to manage global AI risks.

Throughout 2026, thought leaders like Alexandr Wang and organizational analysts stressed the indispensability of systems thinking and adaptive institutional frameworks to govern AI’s accelerating complexity. As Wang notes, the abstraction layer in AI governance is shifting from coding to orchestrating millions of AI agents, necessitating flexible, calibration-based governance that treats autonomy as a dial adjustable to risk and trust levels. This shift demands leadership that embraces uncertainty, fosters distributed decision-making, and builds organizations geared toward scalable learning, all while grounding governance in a deeper philosophical compass to navigate AI’s profound societal transformations.

Sources

Safety Gaps and Dual-Use Dangers

Rapid AI rollouts and dual-use threats expose regulatory blind spots, as safety failures and cyberattacks outpace voluntary industry standards and fragmented state laws.

By mid-2025, pioneering state laws like California’s SB 53 and New York’s RAISE Act began addressing frontier AI risks—such as catastrophic cyberattacks and biothreats—through entity-based regulation that mandates transparency and accountability without stifling innovation. However, these laws risk creating a narrow focus on specific threats like CBRN risks, potentially overlooking the broader and evolving spectrum of AI safety challenges impacting the public today.

The rapid pace of AI and robotics development compresses traditional safety learning curves dramatically, with products moving from prototype to consumer deployment in as little as eighteen months. Despite clear evidence of safety failures—such as large language models endorsing harmful actions against disabled users—and vulnerable populations like nursing home residents facing disproportionate risks, industry often treats safety as a barrier to progress rather than a foundational necessity, underscoring the urgent need for aviation- and medicine-level safety standards including independent certification and mandatory incident reporting.

The dual-use nature of AI is starkly illustrated by incidents like Anthropic’s disruption of a Chinese state-sponsored AI-led cyber espionage campaign, which exposed critical vulnerabilities in models such as Claude through jailbreaking and impersonation attacks. This event, alongside findings of universal jailbreaks across tested AIs and the narrowing capability gap between proprietary and open-source models, highlights the urgent need for robust, continuous, and integrated safety measures that can adapt to rapidly evolving AI-enabled cyber threats beyond voluntary industry commitments.

By early 2026, AI’s capabilities in biology and chemistry had surpassed PhD-level expertise, enabling automation of complex protocols that raise alarming dual-use risks including bioweapon development, while cyberthreat-relevant AI tasks doubled approximately every eight months. Experts from OpenAI warn of imminent, potentially world-shaking cyber and biosecurity threats that demand a resilience-style approach involving coordinated efforts across companies, governments, and cybersecurity platforms, as current regulatory frameworks remain inadequate to address emerging risks like deepfakes and critical infrastructure vulnerabilities such as US water systems.

Sources
AxiosHyperdimensionalFuture TenseTech XploreControlAIAI Safety Newsletter

Get the stories behind the trends

Deep-dive reporting and the weekly brief, in your inbox.