Industrial Policy
for the Intelligence Age:
Ideas to Keep People First
April 2026
Let’s Talk
The drive to understand has always powered human progress—creating a
flywheel from science to technology, from technology to discovery, and from
discovery onward to more science. That inexorable forward movement led us
to melt sand, add impurities, structure it with atomic precision into computer
chips, run energy through those chips, and build systems capable of creating
increasingly powerful artificial intelligence.
In just a few years, AI has progressed from systems capable of fast, narrow tasks to models that can
perform general tasks people used to need hours to do. Now, we’re beginning a transition toward
superintelligence: AI systems capable of outperforming the smartest humans even when they are
assisted by AI. No one knows exactly how this transition will unfold. At OpenAI, we believe we should
navigate it through a democratic process that gives people real power to shape the AI future they want,
and prepare for a range of possible outcomes while building the capacity to adapt. That’s what this
document is for—to start a conversation about governing advanced AI in ways that keep people first.
The promise of superintelligence is extraordinary. Just as electricity transformed homes, the combustion
engine remade mobility, and mass production lowered the cost of essential goods, superintelligence will
speed up scientific and medical breakthroughs, significantly increase productivity, lower costs for
families by making essential goods cheaper, and open the way for entirely new forms of work, creativity,
and entrepreneurship.
Today, AI’s impact on work is often measured by the time required for tasks that systems can reliably
complete. Frontier systems have advanced from supporting tasks that take people minutes to
complete, to tasks that take them hours to complete. If progress continues, we can expect systems to
be capable of carrying out projects that currently take people months. This shift will reshape how
organizations run, how knowledge is created, and how people find meaning and opportunity. It will also
highlight the limitations of today’s policy toolkit and the need for more ambitious ideas to keep people at
the center of the transition to superintelligence.
2
While we strongly believe that AI’s benefits will far outweigh its challenges, we are clear-eyed about the
risks—of jobs and entire industries being disrupted; bad actors misusing the technology; misaligned
systems evading human control; governments or institutions deploying AI in ways that undermine
democratic values; and power and wealth becoming more concentrated instead of more widely shared.
Indeed, we highlight these risks here to raise awareness of the need for policy solutions to address
them. Unless policy keeps pace with technological change, the institutions and safety nets needed to
navigate this transition could fall behind. Ensuring that AI expands access, agency, and opportunity is a
central challenge as we move towards superintelligence. We should aim for a future where
superintelligence benefits everyone, and where we:
1. Share prosperity broadly. The promise of advanced AI is not just technological progress, but a
higher quality of life for all. Everyone should have the opportunity to participate in the new
opportunities AI creates. Living standards should rise and people should see material improvements
through lower costs, better health and education, and more security and opportunity. If AI winds up
controlled by, and benefiting only a few, while most people lack agency and access to AI-driven
opportunity, we will have failed to deliver on its promise.
2. Mitigate risks. The transition toward superintelligence will come with serious risks—from economic
disruption, to misuse in areas like cybersecurity and biology, to the loss of alignment or control over
increasingly powerful systems. Without effective mitigation, people will be harmed. Avoiding these
outcomes requires building new institutions, technical safeguards, and governance frameworks so
that advanced systems remain safe, controllable, and aligned—reducing the risk of large-scale
harm, protecting critical systems, and ensuring people can rely on AI in their daily lives. As capability
scales, safety must scale with it.
3. Democratize access and agency. As capabilities advance, some systems may need to be
controlled for safety. But broad participation in the AI economy should not depend on access to the
most powerful models—it should depend on access to AI that is useful, affordable, preserves
people’s privacy and expands their individual agency. Avoiding a concentration of wealth and
control will require ensuring that people everywhere can use AI in ways that give them real influence
at work, in markets, and through democratic processes.
The Case for a New Industrial Policy. Society has navigated major technological transitions before,
but not without real disruption and dislocation along the way. While those transitions ultimately created
more prosperity, they required proactive political choices to ensure that growth translated into broader
opportunity and greater security. For example, following the transition to the Industrial Age, the
Progressive Era and the New Deal helped modernize the social contract for a world reshaped by
electricity, the combustion engine, and mass production. They did so by building new public
institutions, protections, and expectations about what a fair economy should provide, including labor
protections, safety standards, social safety nets, and expanded access to education.
History shows that democratic societies can respond to technological upheaval with ambition:
reimagining the social contract, mediating between capital and labor, and encouraging broad
3
distribution of the benefits of technological progress while preserving pluralism, constitutional checks
and balances, and freedom to innovate. The transition to superintelligence will require an even more
ambitious form of industrial policy, one that reflects the ability of democratic societies to act collectively,
at scale, to shape their economic future so that superintelligence benefits everyone.
On this path to superintelligence, there are clear steps we need to take today. People are already
concerned about what AI will mean for their lives—whether their jobs and families will be safe, and
whether data centers will disrupt their communities and raise energy prices. AI data centers should pay
their own way on energy so that households aren’t subsidizing them; and they should generate local
jobs and tax revenue. Governments should implement common-sense AI regulation—not to entrench
incumbents through regulatory capture but to protect children, mitigate national security risks, and
encourage innovation.
But the magnitude of the changes we expect and the potential risks we foresee demand even more.
We are entering a new phase of economic and social organization that will fundamentally reshape work,
knowledge, and production. It requires not just incremental policy responses but ambitious policy ideas
for tomorrow that we must start discussing today. This is the moment to start the conversation: to think
boldly, explore new ideas, and collaboratively develop a new industrial policy agenda that ensures
superintelligence benefits everyone.
In normal times, the case for letting markets work on their own is strong. Historically, competition,
entrepreneurship, and open economic participation have lifted living standards and expanded
opportunity. Capitalism, imperfect as it is, remains an effective system for translating human ingenuity
into shared prosperity.
But industrial policy can play an important role when market forces alone aren’t sufficient—when new
technologies create opportunities and risks that existing institutions aren’t equipped to manage. It can
help translate scientific breakthroughs into scaled industries and broad-based economic growth.
A new industrial policy agenda should use government's existing toolbox for aligning public and private
activities: research funding, workforce development, market-shaping tools, and targeted regulation. But
governments should not act alone. Nongovernmental institutions should pilot new approaches,
measure what works, and iterate quickly, then governments should reinforce successes by aligning
incentives and scaling what works through procurement, regulation, and investment. This public-private
collaboration should stave off regulatory capture and centralized control, instead preserving the freedom
to innovate while ensuring that the onset of superintelligence isn’t dominated by the most powerful
forces in society.
We don’t have all, or even most of the answers. Different paths will require different policy responses,
and no single set of tools will be enough in any scenario. But we should aim to build an AI economy
that is both open and resilient through policies that expand participation, broaden access to
opportunity, and ensure that society has the safeguards and institutions needed to manage risk.
4
This document offers initial ideas for an industrial policy agenda to keep people first during the transition
to superintelligence. It is organized in two sections: 1) building an open economy with broad access,
participation, and shared prosperity; and 2) building a resilient society through accountability, alignment,
and management of frontier risks. OpenAI is offering these ideas to help start a broader conversation
about the kinds of policies and institutions needed to navigate the transition, a conversation that needs
to happen among governments, companies, civil society, communities, and families. These ideas are
intentionally early and exploratory, offered not as a comprehensive or final set of recommendations, but
as a starting point for discussion that we invite others to build on, refine, challenge, or choose among
through the democratic process. They also focus on the United States as a starting point, but the
conversation—and the solutions—must ultimately be global. The transition to superintelligence is not a
distant possibility—it’s already underway, and the choices we make in the near term will shape how its
benefits and risks are distributed for decades to come.
1. Building an Open Economy
The promise of advanced AI is that it can benefit everyone by translating abundant intelligence into
extraordinary progress. It can lower the cost of essential goods, expand opportunity, and give people
more time for what is meaningful, relational, and community-building. It can help solve scientific
challenges that still elude human effort: curing or preventing diseases, alleviating food scarcity,
strengthening agriculture under climate stress, and speeding up breakthroughs in clean, reliable energy.
The benefits of major investments in science could emerge within a single lifetime and reach
communities far beyond traditional research hubs.
Yet the same capabilities making this progress possible will also disrupt jobs and reshape entire
industries at a speed and scale unlike any previous technological shift. Some jobs will disappear, others
will evolve, and entirely new forms of work will emerge as organizations learn how to deploy advanced
AI.
These changes will not arrive evenly. Without thoughtful policies, AI could widen inequality by
compounding advantages for those already positioned to capture the upside while communities that
begin with fewer resources fall further behind, excluded from new tools, new industries, and new
opportunities. There is also a risk that the economic gains concentrate within a small number of firms
like OpenAI, even as the technology itself becomes more powerful and widely used. Workers using AI
might well agree that it’s increasing their productivity without believing they’re seeing the benefits.
Maintaining an open economy that is easily accessed and participatory will require ambitious
policymaking. The enclosed ideas include proposals to ensure that workers have a voice in the AI
transition, since workers have deep knowledge about how work is actually performed and where AI can
make work better and safer. Other proposals suggest new mechanisms to share returns from AI-driven
growth by expanding access to capital, sharing economic gains more widely, and aligning the benefits
of AI-enabled growth with higher living standards. And they aim to modernize economic security by
5
helping people navigate transitions, access new opportunities, and maintain stability as work changes.
Together, they form a portfolio of ambitious, preliminary ideas for navigating a wide range of economic
scenarios that the transition to superintelligence might create—all while striving to keep the economy
open and broadly beneficial.
Worker perspectives. Give workers a voice in the AI transition to make work better and safer, including a
formal way to collaborate with management to make sure AI improves job quality, enhances safety, and
respects labor rights. Workers have deep knowledge about how work is actually performed and where
AI can improve outcomes. They will be critical voices in understanding how AI can be used in
workplaces to ensure that technological change will not only lead to improved productivity, but also lead
to better jobs and stronger, safer workplaces. Allow workers to prioritize AI deployments that improve
job quality by eliminating dangerous, repetitive, administrative, or exhausting tasks so employees can
focus on higher-value work. At the same time, set clear limits on harmful uses of AI that could erode job
quality by intensifying workloads, narrowing autonomy, or undermining fair scheduling and pay.
AI-first entrepreneurs. Help workers turn domain expertise into new companies by using AI to handle
the overhead that usually blocks entrepreneurship (e.g., accounting, marketing, procurement). Pair
microgrants or revenue-based financing with practical “startup-in-a-box” supports such as model
contracts and shared back-office infrastructure so that new small businesses can compete quickly.
Worker organizations could serve as enablers by offering training, providing shared services, and
helping workers negotiate fair commercial terms and protect IP.
Right to AI. Treat access to AI as foundational for participation in the modern economy, similar to mass
efforts to increase global literacy, or to make sure that electricity and the internet reach remote parts of
the globe. (The internet still isn’t fairly deployed across the globe or even the US; learn from this and
seek to rectify those issues when it comes to AI.) Expand affordable, reliable access to foundational
models—the building blocks of modern AI systems—and make a baseline level of capability broadly
available, including through free or low-cost access points. Support the education, infrastructure,
connectivity, and training needed to use these systems effectively, and make sure that workers, small
businesses, schools, libraries, and underserved communities are not excluded from the capabilities that
drive productivity and opportunity.
Modernize the tax base. As AI reshapes work and production, the composition of economic activity
may shift—expanding corporate profits and capital gains while potentially reducing reliance on labor
income and payroll taxes. This could erode the tax base that funds core programs like Social Security,
Medicaid, SNAP, and housing assistance—putting them at risk. Tax policy should adapt to ensure these
systems remain durable. Policymakers could rebalance the tax base by increasing reliance on
capital-based revenues—such as higher taxes on capital gains at the top, corporate income, or
targeted measures on sustained AI-driven returns—and by exploring new approaches such as taxes
related to automated labor. These reforms should be paired with wage-linked incentives that encourage
firms to retain, retrain, and invest in workers, similar to existing R&D-style credits. Together, these
changes would help stabilize funding for essential programs while supporting workforce transitions in an
AI-driven economy.
6
Public Wealth Fund. Create a Public Wealth Fund that provides every citizen—including those not
invested in financial markets—with a stake in AI-driven economic growth. While tax reforms help ensure
governments can continue to fund essential programs, a Public Wealth Fund is designed to ensure that
people directly share in the upside of that growth. Policymakers and AI companies should work
together to determine how to best seed the Fund, which could invest in diversified, long-term assets
that capture growth in both AI companies and the broader set of firms adopting and deploying AI.
Returns from the Fund could be distributed directly to citizens, allowing more people to participate
directly in the upside of AI-driven growth, regardless of their starting wealth or access to capital.
Accelerate grid expansion. Establish new public-private partnership models to finance and accelerate
the expansion of energy infrastructure required to power AI. Use these models to address financing
constraints, permitting delays, and siting risks that have limited high-voltage interstate and interregional
transmission—and to deliver infrastructure at speed and scale, limit taxpayer risk, and share the upside
with the public. Approaches could include reducing the cost of capital through targeted investment
credits, direct and indirect flexible subsidies, or equity stakes; removing market barriers to advanced
technologies such as advanced conductors and high voltage direct current; and providing a narrow
federal authority to accelerate the construction of interregional transmission when it is in the national
interest. Partnerships should be structured to minimize taxpayer exposure to commercial losses and
ensure that expanded energy infrastructure translates into lower energy costs for households and
businesses.
Efficiency dividends. Convert efficiency gains from AI into durable improvements in workers’ benefits
when routine workload declines and operating costs fall, including incentivizing companies to increase
retirement matches or contributions, cover a larger share of healthcare costs, and subsidize child and
eldercare. Incentivize employers and unions to run time-bound 32-hour/four-day workweek pilots with
no loss in pay that hold output and service levels constant, then convert reclaimed hours into a
permanent shorter week, bankable paid time off, or both. Where helpful, firms could also offer
predictable “benefits bonuses” tied to measured productivity improvements so the efficiency dividend
shows up both as long-term financial security and as time back for workers.
Adaptive safety nets that work for everyone. Make sure the existing safety net works reliably, quickly,
and at scale, because if the transition to superintelligence is going to benefit everyone, the systems
designed to provide economic and health security need to deliver without delay or gaps. That starts
with unemployment insurance, SNAP, Social Security, Medicaid, and Medicare that are not just in place
but fully functional, accessible, and responsive to the realities people will face during the transition.
Next, invest in clear, real-time measurement of how AI is affecting work, wages, job quality, and sectoral
dynamics, using public metrics such as unemployment rates and indicators of regional or
industry-specific displacement. These systems should provide policymakers with timely visibility into
where disruption is occurring and how severe it is. Then, define a package of temporary, expanded
safety nets (e.g., expanded or more flexible unemployment benefits, fast cash assistance, wage
insurance, training vouchers) that activates automatically when these metrics exceed pre-defined
thresholds. When disruption rises above those levels, support would scale up; as conditions stabilize, it
7
would phase out. This ensures that assistance is targeted, time-bound, and proportional to the scale of
disruption, and also avoids a permanent expansion of programs.
Portable benefits. Over time, build benefit systems that are not tied to a single employer by expanding
access to healthcare, retirement savings, and skills training through portable accounts that follow
individuals across jobs, industries, education programs, and entrepreneurial ventures. Public programs
can decouple key benefits from employment status by expanding access to retirement and training
support regardless of where or how someone works. Implementation can run through portable benefit
platforms that pool contributions from multiple sources and route them into standardized accounts
attached to the individual, not the job. Retirement systems can also be modernized through pooled
structures that allow workers to accrue benefits continuously across employers, reducing gaps and
preserving continuity over time.
Pathways into human-centered work. Expand opportunities in the care and connection
economy—childcare, eldercare, education, healthcare, and community services—as pathways for
workers displaced by AI. Although AI can enhance these roles by reducing administrative burdens and
enabling greater personalization, human connection will remain an essential part of the profession. As AI
reshapes the labor market, these sectors can absorb transitioning workers if supported with
investments in training, wages, and job quality. Governments can build training pipelines, support
transitions into care roles, and incentivize employers to raise pay and improve conditions in fields facing
chronic shortages.
These initiatives could be complemented with a family benefit that recognizes caregiving as
economically valuable work and supports evolving work patterns. This benefit could help cover
childcare, education, and healthcare while remaining compatible with part-time work, retraining, or
entrepreneurship. Together, these efforts would expand access to care, strengthen communities, and
create meaningful, human-centered work.
Accelerate scientific discovery and scale the benefits. Build a distributed network of AI-enabled
laboratories to dramatically expand the capacity to test and validate AI-generated hypotheses at scale.
These labs would integrate AI systems directly into experimental workflows by automating routine
processes, capturing high-quality data, and enabling rapid iteration between hypothesis generation and
testing. Then, build the physical systems and infrastructure needed to translate validated discoveries
into real-world use at scale. This includes expanding the capacity of organizations to deploy new
technologies, upgrading facilities and systems required for implementation, and aligning financing and
incentives to support adoption. It also includes a sustained investment in people: training scientists,
technicians, and operators to contribute to AI-enabled science.
These investments ensure that breakthroughs move beyond laboratories and into widespread use,
while strengthening the workforce and operational systems required to build, maintain, and run the
infrastructure that supports AI-enabled discovery. Both laboratory and production infrastructure should
be deployed broadly across universities, community colleges, hospitals, and regional research hubs, not
concentrated in a small number of elite institutions.
8
2. Building a Resilient Society
As AI systems become more capable and more embedded across the economy, they may introduce
new vulnerabilities alongside new abundance. Some systems may be misused for cyber or biological
harm. Others may create new pressures on social and emotional well-being, including for young people,
if deployed without adequate safeguards. AI systems may act in ways that are misaligned with human
intent or operate beyond meaningful human oversight. And as advanced AI reshapes how people,
organizations, and governments operate, it may place new strain on the institutions and norms that
societies rely on to remain stable, secure, and free.
We should be clear-eyed about the resilience required here. These new risks won’t be isolated or
suitable for addressing one at a time—AI will reshape how work is performed, how decisions are made,
how organizations operate, and how states interact. Building resilience therefore means making sure
people and institutions can adapt quickly, maintain meaningful agency over how these systems are
used, and preserve broadly shared prosperity even as economic and social structures evolve.
Over the past several years, leading AI developers including OpenAI have focused heavily on upstream
safeguards: development of global standards, transparency around evaluations, mitigations, and risks,
and investments in model testing, red teaming, and usage policies designed to identify and mitigate
risks before deployment. Policymakers have also focused here, codifying requirements in the EU AI Act
and in US state-based regulation. At the same time, training and literacy efforts have expanded so that
schools, nonprofits, small businesses, and communities can use AI tools more safely and effectively.
These upstream efforts should continue.
But as AI systems become more capable and more widely deployed, resilience will also depend upon
what happens after deployment—when systems must be monitored in real time, operate under
uncertainty, and integrate into institutions not designed for agentic workflows.
This is not a new challenge. When transformative technologies have reshaped society in the past, they
have introduced new risks alongside new benefits, and new systems were built to manage them as they
scaled. As electricity spread, societies built safety standards and regulatory institutions. As automobiles
transformed mobility, safety systems reduced risk while preserving freedom of movement. In aviation,
continuous monitoring and coordinated response systems made flying one of the safest forms of
transportation. In food and medicine, testing and post-market surveillance helped ensure safety in
everyday use. In each case, resilience was not automatic—it was built with the luxury of time.
As we move toward superintelligence, building a resilient society will require a similar but speedier effort
that kicks into gear now. The ideas below are a slate of ambitious approaches to building a more
resilient society. They focus on building and scaling safety systems that operate in real-world conditions
by establishing mechanisms for trust, accountability, and auditing. They suggest opportunities for
strengthening governance so that advanced AI remains controllable, transparent, and aligned with
democratic values. And they suggest approaches to improve coordination across companies,
governments, and countries so that risks can be identified early, information can be shared, and
9
responses can be executed quickly when needed. Together, these proposals extend important safety
work already underway and represent initial ideas to keep AI safe, governable, and aligned with
democratic values.
Safety systems for emerging risks. Research and develop tools that protect models, detect risks, and
prevent misuse across high-consequence domains, including cyber and biological risks as well as other
pathways to large-scale harm. Expand the use of advanced AI systems for threat modeling, red
teaming, net assessments, and robustness testing to identify and anticipate novel risks early and inform
mitigation strategies. Develop and scale complementary protective systems; for example, rapid
identification and production of medical countermeasures in the event of an outbreak and expanded
strategic stockpiles to prepare for future risks. Then, catalyze competitive safety markets by creating
sustained demand for these capabilities through procurement, standards, insurance frameworks, and
advance-purchase commitments. Over time, this approach can make safeguards an output of
innovation and competition, ensuring that defenses improve as quickly as the risks they are designed to
address.
AI trust stack. Research and develop systems that help people trust and verify AI systems, the content
they produce, and the actions they take—especially as these systems take on more real-world
responsibilities. Advance the development of provenance and verification standards and tools that can
build trust in AI systems while preserving privacy. This could include enabling secure, verifiable
signatures for actions such as generating content or issuing instructions, and developing
privacy-preserving logging and audit systems capable of supporting investigation and accountability
without enabling pervasive surveillance.
These types of solutions should capture key information about system behavior and use while
minimizing the collection of sensitive data, and be designed to support investigation or intervention
under clearly defined legal or safety conditions. This work could also include developing and testing
governance frameworks that clarify responsibility within organizations, including how accountability
could be assigned to specific roles and how delegation, monitoring, and escalation processes could
function as systems become more capable. Over time, these efforts could establish a foundation for
accountability by building trust in AI interactions and helping ensure that when harm occurs,
responsibility can be appropriately allocated.
Auditing regimes. Strengthen institutions such as the Center for AI Standards and Innovation (CAISI) to
develop auditing standards for frontier AI risks in coordination with national security agencies. Use tools
such as government procurement, advance-purchase commitments, insurance frameworks, and
standards-setting to create and scale a competitive market of auditors and evaluators capable of
assessing AI systems and products for safety and security risks, building auditing capacity alongside
the technology. Standards should be designed for international adoption to reduce fragmentation and
avoid creating unnecessary compliance burdens for small companies, as well as those operating across
jurisdictions.
10
As we progress toward superintelligence, there may come a point where a narrow set of highly capable
models—particularly those that could materially advance chemical, biological, radiological, nuclear, or
cyber risks—require stronger controls, including pre- and post-deployment audits using the standards
developed in advance. Apply these requirements only to a small number of companies and the most
advanced models, preserving a vibrant ecosystem of less powerful systems and the startups building
on them. This approach maintains broad access to general-purpose AI while applying targeted
safeguards where failures could create the greatest harm, avoiding unnecessary barriers that could limit
competition or enable regulatory capture.
Model-containment playbooks. Develop and test coordinated playbooks to contain dangerous AI
systems once they have been released into the world. As AI capabilities advance, societies may face
scenarios where dangerous systems cannot be easily recalled—because model weights have been
released, developers are unwilling or unable to limit access to dangerous capabilities, or the systems
are autonomous and capable of replicating themselves. In these cases, the challenge is containment:
limiting the spread of dangerous capabilities, reducing harm, and coordinating responses under
real-world constraints. Experience from other high-consequence domains, such as cybersecurity and
public health, shows that even when full containment is not possible, coordinated action can still
meaningfully reduce impact.
Mission-aligned corporate governance. Frontier AI companies should adopt governance structures that
embed public-interest accountability into decision-making, such as Public Benefit Corporations with
mission-aligned governance. These structures should include explicit commitments to ensure that the
benefits of AI are broadly shared, including through significant, long-term philanthropic or charitable
giving. At the same time, harden frontier systems against corporate or insider capture by securing
model weights and training infrastructure, auditing models for manipulative behaviors or hidden loyalties,
and monitoring high-risk deployments so no individual or internal faction can quietly use AI systems to
concentrate power.
Guardrails for government use. Have policymakers establish clear rules for how governments can and
cannot use AI, with especially high standards for reliability, alignment, and safety. These standards
should be codified in law and reinforced through technical safeguards. At the same time, use AI to
strengthen democratic accountability. As more government decisions are made through AI-assisted
workflows, these systems will create clearer digital records of government reasoning and action that
can be logged alongside other public records. With appropriate safeguards, oversight institutions such
as inspectors general, congressional committees, and courts could use AI-enabled auditing tools to
detect abuse, identify harms, and improve accountability at scale.
Also, modernize transparency frameworks (including the Freedom of Information Act) to allow citizens
and watchdog organizations to use AI to review targeted questions about government actions while
protecting sensitive information. This could include clarifying when AI-interaction logs and agentic action
logs constitute federal records that must be retained for specified periods.
11
Mechanisms for public input. Create structured ways for public input so that alignment isn’t defined only
by engineers or executives behind closed doors. As advanced AI makes more decisions that affect
people’s lives, societies need shared clarity about what these systems are supposed to do, what values
should guide them, and how well they are performing. Make alignment more democratic, legible, and
accountable through transparent specifications, evaluation frameworks, and representative input
processes. Developers should publish model specifications that describe how systems are intended to
behave and share information about how those systems are evaluated. Governments and public
institutions should help shape these standards by anchoring them in democratic laws and values, while
establishing mechanisms for representative public input to be considered alongside traditional business
stakeholders. Together, these approaches help ensure that the advancement of AI reflects the
perspectives of the societies that must live with its consequences.
Incident reporting. Establish a mechanism for companies to share information about incidents, misuse,
and near-misses with a designated public authority. The system should emphasize learning and
prevention over punishment, with appropriately scoped public disclosures that ensure transparency and
democratic oversight while protecting sensitive technical, national security, and competitive information.
Near-miss reporting could include cases where models exhibited concerning internal reasoning,
unexpected capabilities, or other warning signals—even if safeguards ultimately prevented harm—so
the ecosystem can learn from close calls before they become real incidents.
International information-sharing around AI capabilities, risks, and mitigations. Strengthen national
evaluation institutions as the foundation for international coordination, beginning with expanding the role
of the CAISI as a trusted technical body for evaluating frontier systems, assessing safeguards, and
informing government understanding of advanced AI capabilities. Building on this foundation, develop a
global network of AI Institutes that collaborate through shared protocols for information exchange, joint
evaluations, and coordinated mitigation measures.
Over time, this network could evolve into an international framework akin to the other multilateral
institutions focused on safety and standards, one that gives trusted public authorities visibility into
frontier AI development; and creates secure cross-lab and cross-country channels for sharing
evaluation results, alignment findings, and emerging risks; and likewise supports communicating during
crises. To enable effective collaboration, policymakers should ensure that companies can share safety-
and risk-related information through these channels without running afoul of antitrust or competition
constraints, using clear safe harbors and narrowly scoped information-sharing rules. This system should
expand beyond a narrow focus on national security to include a broader range of societal risks,
including impacts on youth safety and well-being.
12
Starting the Conversation
We offer these ideas not as fixed answers but as a starting point for a broader conversation about how
to ensure that AI benefits everyone. That conversation should be inclusive and ongoing—engaging
governments, companies, researchers, civil society, communities, and families—and should be
mediated through democratic processes that give people real power to shape the AI future they want. It
also needs to expand globally—bringing in the perspectives of cultures, societies, and governments
around the world.
These ideas are our first contribution to that effort, but only the beginning. Progress will depend on
continued iteration, experimentation, and collaboration across institutions and sectors. To help sustain
momentum, OpenAI is: (1) welcoming and organizing feedback through
newindustrialpolicy@openai.com; (2) establishing a pilot program of fellowships and focused research
grants of up to $100,000 and up to $1 million in API credits for work that builds on these and related
policy ideas; and (3) convening discussions at our new OpenAI Workshop opening in May in
Washington, DC.
13