Require frontier AI labs to conduct nearly all R&D publicly but keep the weights closed

ai-regulation open-science transparency research-policy open-source ai-policy ai-safety ai-governance

Share this with your representatives, journalists and friends

Key insights

Both sides often separate development transparency from weight release, but they disagree on whether closing weights is the prudent line or a major loss of openness (18356, 17401, 17437, 17451).
Opponents frequently frame openness as both an innovation strategy and a safety mechanism, not just a philosophical preference (19246, 17460, 17443, 17442).
Supporters concentrate their case on governance and misuse prevention: public research can let outsiders scrutinize labs, while closed weights may limit rapid, irreversible spread of harmful capabilities (19286, 18356, 19226).
A recurring divide is whether transparency should be public for everyone or targeted to governments, peer labs, or trusted evaluators (17410, 17444, 19447, 17401).
Across the quotes, open-source organizations, public-interest groups, universities, and labs repeatedly treat weights, data, code, and methods as a single reproducibility and accountability package (17437, 17430, 17425, 17450).

Arguments for

Governance through transparency

Proponents say labs should be transparent about everything except model weights so outside groups can check their homework. They argue that total research transparency would simplify both writing governance rules and enforcing them.

AI companies should be transparent about everything but the model weights, so outside groups can check the AI company's homework.

Daniel Kokotajlo · Jul 9, 2026 · full quote
Total research transparency would fundamentally and radically simplify the task of governing AI alignment: both figuring out what the rules should be in the first place, and then actually enforcing them.

Ajeya Cotra · Jul 10, 2026 · full quote

Contain weight-specific risks

Supporters argue that open-weight models pose unique risks because harmful capabilities can proliferate rapidly and irreversibly, and safeguards can be removed quickly and cheaply. In this view, releasing algorithms or other research need not imply releasing weights.

Open-weight systems are the lifeblood of open research and innovation in AI. They increase transparency, allow for widespread red teaming, and decrease market concentration. However, open-weight models also pose unique risks. They can allow harmful AI capabilities to proliferate …

UK AI Security Institute · Aug 25, 2026 · full quote

Open methods, controlled access

Another pro case envisions open algorithms and extensive development transparency, but non-public weights and controlled model access. Its proponents argue that public evaluations or fine-tuning under transparent conditions can capture most of openness's benefits without turning weight access into unrestricted deployment.

Plan A is not that crazy, but it’s close: the algorithms are mandated to be open source, and while the weights aren’t (in fact the weights are banned from being OS), access to the weights is mandated to be open; members of the public can do evaluations (including fine-tuning!) on…

AI Futures Project · Jul 9, 2026 · full quote
AI companies should be transparent about everything but the model weights, so outside groups can check the AI company's homework.

Daniel Kokotajlo · Jul 9, 2026 · full quote

Arguments against

Innovation and broad access

Opponents argue that openness is foundational to AI progress and that broad access to models, data, and tools keeps AI in more people's hands. They say sharing code and models helps academia, nonprofits, and others participate in progress rather than leaving it to a few closed labs.

If I were to try to give one prescription, it would be to promote openness, because AI is amazing technology and I want it to be in everyone’s hands.

Andrew Ng · Aug 12, 2026 · full quote
The best AI tools in the world won’t be built behind closed doors by a handful of companies. They’ll be built by people who ship code, share it, and let others make it better.

Kanishka Narayan · Jun 10, 2026 · full quote
Open innovation is foundational to AI progress, and broad access to models, data, and tools is essential to accelerating breakthroughs. By adopting the Linux Foundation’s OpenMDW framework for NVIDIA open model families, we’re helping establish a simpler, more consistent standard…

Kari Briski · May 28, 2026 · full quote
These days you might argue that progress in AI is getting limited into a few closed labs, but it’s actually very important to the vast majority of academia and researchers, or nonprofit and other places who want to also be part of this progress. And we’ve seen that all this progr…

Hanna Hajishirzi · Mar 25, 2026 · full quote
We believe that open-wide models should be the basis for building all the AI software in the world. By having an open ecosystem of people that have aligned incentives to create assets that are going to be great for humanity, we can accelerate progress and make sure that everybody…

Arthur Mensch · Mar 25, 2026 · full quote
The biggest threat to AI innovation is the closed nature of the practice. We have been pushing very, very strongly towards openness.

Ali Farhadi · Feb 19, 2025 · full quote

Concentration and public interest

Critics also argue that open source prevents centralization and keeps frontier systems from becoming closed boxes held by a few companies. They link openness to competitiveness, public-interest infrastructure, and making AI useful and open to the world rather than the few.

Open source is a positive and important force for empowering people and preventing centralization that is detrimental for both safety and the economy. Meta continues to be strongly supportive of open source, including open source AI models. The current open source ecosystem is st…

Mark Zuckerberg · Aug 10, 2026 · full quote
We are extremely interested in looking at ways to build U.S. open source, make it competitive, make it the preferential adoption by planet Earth,

Sean Cairncross · Aug 5, 2026 · full quote
We publish weights, training code and synthetic-data pipelines wherever the research permits. Apache 2.0 is the default. We don't believe the systems on the path to general intelligence should be closed boxes held by three companies.

Nuro AI Labs · 2026 · full quote
For that to happen, the building blocks of AI—the data that powers these models and the models themselves—must serve and be shaped by the public interest. [...] a rare and exciting opportunity for open, transparent, and accountable AI infrastructure to gain ground.

Martin Tisné · May 16, 2025 · full quote
We’re a research group pushing the frontier of artificial intelligence — and making it useful, affordable, and open to the world. Not the few. Everyone. We do original research at the capability frontier — and publish what we learn instead of locking it away.

Frontier Intelligence Labs · 2026 · full quote

Openness as safety asset

A major counterargument is that open models and open tooling are defensive assets, enabling independent evaluation, broader community assessments, shared remediation, trust, auditability, and stronger safety certifications. On this view, openness is not a security liability but part of how risks are anticipated and mitigated.

As policymakers and regulators work through questions of AI safety, the distinction that matters is not open versus closed, but transparent and secure versus opaque and unexamined. Open models and open tooling should be recognized as defensive assets that enable transparency, ind…

The Linux Foundation · Jul 27, 2026 · full quote
Our open source approach also helps us to better anticipate and mitigate risk because it enables us to learn from the broader community’s independent assessments of our models’ capabilities.

Meta · Feb 3, 2025 · full quote
Fully transparent models providing access to their training data and procedures can support the most extensive safety certifications.

Hugging Face · Mar 19, 2025 · full quote
We believe transparency and open research are not optional — they are the only way to build AI that people can trust, audit, and improve upon. Tribela Labs exists to prove that a sustainable company can advance AI research while keeping its methods transparent and its commitments…

Tribela Labs · 2026 · full quote
Mozilla has long advocated for such rules to be included in the AI Act to ensure that large AI labs must play their part in making the technology they develop safer and more transparent.

Maximilian Gahntz · Aug 11, 2025 · full quote

Full-pipeline accountability

Many opponents reject the idea that public research plus closed weights would count as real openness. They argue that open source AI requires access to trained parameters alongside code, data, methods, and documentation so the full pipeline can be scrutinized, validated, reproduced, and improved.

Open source AI requires: unrestricted access to trained parameters/weights; complete code used to train the system and prepare training data; full dataset (unless legally impossible).

Stefano Maffulli · Oct 6, 2025 · full quote
For AI to be both accountable and scalable, every part of the pipeline—from the initial dataset to the final set of parameters—needs to be open to scrutiny, validation, and collective improvement.

Open Source Initiative · Jan 29, 2025 · full quote
The entire development process, including its architecture, model weights, training data and methods, is openly accessible and fully documented.

ETH Zurich · Sep 2, 2025 · full quote
Fully open, end-to-end recipe for language model training: [...] all ingredients openly available: data, training recipe, methodology, and software.

Ai2 (Allen Institute for AI) · Mar 13, 2025 · full quote
All mixtures, filtering rules, and data sources are fully released for reproducibility.

Institute of Foundation Models (IFM) · Dec 1, 2025 · full quote
Open weights are progress. You can download the model, run it on your own machine, keep it out of someone else’s data pipeline. But you still can’t see how the thing was built, what it was trained on, or why it behaves the way it does. That’s not an open model. That’s open distri…

James Landay · Aug 4, 2026 · full quote

Middle ground

Targeted disclosure channels

Some quotes support transparency obligations, but narrower ones: disclosures to government, other labs, or trusted evaluators, or public visibility focused on safety practices rather than all research. They present this as a practical first step that preserves private-sector agility and avoids requiring labs to reveal everything to the world.

The research community, while they should not be expected to reveal to the world where they have got to, should be in dialogue with government. They need to be required to tell government where they are and what they are doing on the research front.

Baroness Neville-Jones · Jan 29, 2026 · full quote
Internal model transparency is a minimum viable form of research transparency: instead of sharing every innovation, labs share only the finished artifact – the model itself. And while AI 2040 argues for this transparency to be total, to all members of the public, I suggest the mo…

Karthik Tadepalli · Sep 1, 2026 · full quote
Our proposed transparency framework offers a practical first step: public visibility into safety practices while preserving private sector agility to deliver AI's transformative potential.

Anthropic · Jul 7, 2025 · full quote

Open by default exceptions

Another nuanced view rejects a simple open-versus-closed dichotomy and favors publishing research, code, data, or weights as widely as can be shown to be safe, while protecting what needs protecting and documenting any withholding. This approach treats openness and restraint as case-by-case decisions rather than fixed absolutes.

It’s very dangerous to make this a dichotomy between complete openness all the way to complete closedness. In complex software systems as well as scientific systems it’s much more nuanced.

Fei-Fei Li · Aug 12, 2026 · full quote
We’re open by default: open code; open data, subject to privacy and copyright; and open weights wherever responsible. We release as widely as we can demonstrate to be safe.

Tertium AI · 2026 · full quote
Research, code, weights, datasets, evaluations, and governance artifacts default to open. Any decision to withhold is published with its rationale, its scope, and the conditions under which it will be revisited.

Tenzro Foundation · 2026 · full quote
We believe transparency and open research are not optional — they are the only way to build AI that people can trust, audit, and improve upon. Tribela Labs exists to prove that a sustainable company can advance AI research while keeping its methods transparent and its commitments…

Tribela Labs · 2026 · full quote

The debate is less about whether transparency matters than about what must be opened, to whom, and under what constraints (18356, 17444, 17451). One camp argues that near-total public R&D with closed weights offers a workable balance between oversight and misuse prevention, while the other argues that closing weights would undercut the innovation, accountability, and safety benefits they associate with genuine openness (19226, 18356, 19286, 17460, 17437, 17451). A further set of quotes favors narrower or conditional disclosure, such as sharing with government, peer labs, or trusted evaluators, or opening by default with explicit exceptions (17410, 17444, 19447, 19069, 17461).

Cast your vote:

Results (48 votes):

Total (48 votes)

For 6 (13%) Abstain 0 (0%) Against 42 (88%)

For (6)

UK AI Security Institute
United Kingdom government institute for evaluating advanced AI systems.

I am this person! Add a sourced quote Report comment

votes For and says:

Open-weight systems are the lifeblood of open research and innovation in AI. They increase transparency, allow for widespread red teaming, and decrease market concentration. However, open-weight models also pose unique risks. They can allow harmful A...
more AI Verified source (Aug 25, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Fei-Fei Li
Stanford AI professor; HAI co-director

I am this person! Add a sourced quote Report comment

votes For and says:

It’s very dangerous to make this a dichotomy between complete openness all the way to complete closedness. In complex software systems as well as scientific systems it’s much more nuanced.
AI Verified source (Aug 12, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Thomas Larsen
AI Futures Project researcher writing on Plan A assumptions, AI governance, and AI safety strategy.

I am this person! Add a sourced quote Report comment

votes For and says:

Proposal #1: Total Research Transparency. Nearly all AI research is fully transparent to the public, including AI algorithms, code, and documentation, for all frontier AI projects. AI model weights, significant fractions of the training data, and a s...
more AI Verified source (Jul 13, 2026)

added 9d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Daniel Kokotajlo
AI Futures Project founder

I am this person! Add a sourced quote Report comment

votes For and says:

AI companies should be transparent about everything but the model weights, so outside groups can check the AI company's homework.
AI Verified source (Jul 9, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
AI Futures Project
Nonprofit AI forecasting research group behind AI 2027 and Plan A; led by Daniel Kokotajlo.

I am this person! Add a sourced quote Report comment

votes For and says:

Plan A is not that crazy, but it’s close: the algorithms are mandated to be open source, and while the weights aren’t (in fact the weights are banned from being OS), access to the weights is mandated to be open; members of the public can do evaluatio...
more AI Verified source (Jul 9, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate

Loading more opinions...

Abstain (0)

Loading more opinions...

Against (42)

Karthik Tadepalli
Research Scholar at the Centre for the Governance of AI, studying the economic effects of transformative AI; PhD in Economics from UC Berkeley.

I am this person! Add a sourced quote Report comment

votes Against and says:

Internal model transparency is a minimum viable form of research transparency: instead of sharing every innovation, labs share only the finished artifact – the model itself. And while AI 2040 argues for this transparency to be total, to all members o...
more AI Verified source (Sep 1, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Andrew Ng
Baidu; Stanford CS faculty; founded Coursera and Google Brain

I am this person! Add a sourced quote Report comment

votes Against and says:

If I were to try to give one prescription, it would be to promote openness, because AI is amazing technology and I want it to be in everyone’s hands.
AI Verified source (Aug 12, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Mark Zuckerberg
Meta CEO; Facebook co-founder

I am this person! Add a sourced quote Report comment

votes Against and says:

Open source is a positive and important force for empowering people and preventing centralization that is detrimental for both safety and the economy. Meta continues to be strongly supportive of open source, including open source AI models. The curre...
more AI Verified source (Aug 10, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Sean Cairncross
U.S. National Cyber Director

I am this person! Add a sourced quote Report comment

votes Against and says:

We are extremely interested in looking at ways to build U.S. open source, make it competitive, make it the preferential adoption by planet Earth,
AI Verified source (Aug 5, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
James Landay
Denning Director of Stanford’s Institute for Human-Centered Artificial Intelligence (Stanford HAI).

I am this person! Add a sourced quote Report comment

votes Against and says:

Open weights are progress. You can download the model, run it on your own machine, keep it out of someone else’s data pipeline. But you still can’t see how the thing was built, what it was trained on, or why it behaves the way it does. That’s not an ...
more AI Verified source (Aug 4, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
The Linux Foundation
Nonprofit consortium supporting open-source technology ecosystems

I am this person! Add a sourced quote Report comment

votes Against and says:

As policymakers and regulators work through questions of AI safety, the distinction that matters is not open versus closed, but transparent and secure versus opaque and unexamined. Open models and open tooling should be recognized as defensive assets...
more AI Verified source (Jul 27, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Jakob Foerster
Associate professor at the University of Oxford and co-director of the British Open-ended Learning and Discovery Lab

I am this person! Add a sourced quote Report comment

votes Against and says:

BOLD is about a different route: discovering fundamentally new ways to build AI that are more efficient, more open and better aligned with human needs.
AI Verified source (Jun 23, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Vince McMullin
Co-founder and CEO of CanXP AI, a Canadian sovereign-AI platform.

I am this person! Add a sourced quote Report comment

votes Against and says:

Release the Fable 5 weights under a responsible open-weight framework. Provide meaningful training-corpus disclosure. Create audited access pathways for independent researchers and sovereign AI operators. Allow the public to verify safety claims rath...
more AI Verified source (Jun 11, 2026)

added 24d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Kanishka Narayan
UK Minister for AI and Online Safety; Parliamentary Under-Secretary of State, Department for Science, Innovation and Technology

I am this person! Add a sourced quote Report comment

votes Against and says:

The best AI tools in the world won’t be built behind closed doors by a handful of companies. They’ll be built by people who ship code, share it, and let others make it better.
AI Verified source (Jun 10, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Kari Briski
Vice president of generative AI software at NVIDIA

I am this person! Add a sourced quote Report comment

votes Against and says:

Open innovation is foundational to AI progress, and broad access to models, data, and tools is essential to accelerating breakthroughs. By adopting the Linux Foundation’s OpenMDW framework for NVIDIA open model families, we’re helping establish a sim...
more AI Verified source (May 28, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Hanna Hajishirzi
Professor at the University of Washington and senior director of NLP at Ai2

I am this person! Add a sourced quote Report comment

votes Against and says:

These days you might argue that progress in AI is getting limited into a few closed labs, but it’s actually very important to the vast majority of academia and researchers, or nonprofit and other places who want to also be part of this progress. And ...
more AI Verified source (Mar 25, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Arthur Mensch
Mistral AI co-founder and CEO

I am this person! Add a sourced quote Report comment

votes Against and says:

We believe that open-wide models should be the basis for building all the AI software in the world. By having an open ecosystem of people that have aligned incentives to create assets that are going to be great for humanity, we can accelerate progres...
more AI Verified source (Mar 25, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Misha Laskin
Co-founder and CEO of Reflection AI; artificial intelligence researcher

I am this person! Add a sourced quote Report comment

votes Against and says:

Models are fundamental knowledge infrastructure, and fundamental knowledge infrastructure yearns for openness. There’s a flourishing ecosystem of powerful, closed models but equally capable open models that are going to be coming over the next couple...
more AI Verified source (Mar 25, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
John deVadoss
AI technologist; co-founder of NeuralFabric and former Microsoft general manager

I am this person! Add a sourced quote Report comment

votes Against and says:

We cannot meaningfully regulate the development of AI in a world where artifacts copy in near real time and research flows fluidly across borders.
AI Verified source (Feb 2, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Baroness Neville-Jones
British Conservative life peer and former UK security minister.

I am this person! Add a sourced quote Report comment

votes Against and says:

The research community, while they should not be expected to reveal to the world where they have got to, should be in dialogue with government. They need to be required to tell government where they are and what they are doing on the research front.
AI Verified source (Jan 29, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Nuro AI Labs
London-based AI research lab developing open models and general-intelligence research

I am this person! Add a sourced quote Report comment

votes Against and says:

We publish weights, training code and synthetic-data pipelines wherever the research permits. Apache 2.0 is the default. We don't believe the systems on the path to general intelligence should be closed boxes held by three companies.
AI Verified source (2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Tribela Labs
European AI research lab focused on open, transparent, privacy-preserving systems

I am this person! Add a sourced quote Report comment

votes Against and says:

We believe transparency and open research are not optional — they are the only way to build AI that people can trust, audit, and improve upon. Tribela Labs exists to prove that a sustainable company can advance AI research while keeping its methods t...
more AI Verified source (2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Frontier Intelligence Labs
Frontier AI research group founded in 2026 to develop broadly accessible systems

I am this person! Add a sourced quote Report comment

votes Against and says:

We’re a research group pushing the frontier of artificial intelligence — and making it useful, affordable, and open to the world. Not the few. Everyone. We do original research at the capability frontier — and publish what we learn instead of locking...
more AI Verified source (2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Tertium AI
Nonprofit building and distributing frontier AI with open-by-default governance

I am this person! Add a sourced quote Report comment

votes Against and says:

We’re open by default: open code; open data, subject to privacy and copyright; and open weights wherever responsible. We release as widely as we can demonstrate to be safe.
AI Verified source (2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Tenzro Foundation
Public-interest foundation developing open and accountable AI infrastructure

I am this person! Add a sourced quote Report comment

votes Against and says:

Research, code, weights, datasets, evaluations, and governance artifacts default to open. Any decision to withhold is published with its rationale, its scope, and the conditions under which it will be revisited.
AI Verified source (2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Institute of Foundation Models (IFM)
AI research institute behind the fully open LLM360 model program

I am this person! Add a sourced quote Report comment

votes Against and says:

All mixtures, filtering rules, and data sources are fully released for reproducibility.
AI Verified source (Dec 1, 2025)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Stella Biderman
EleutherAI executive director and scientist

I am this person! Add a sourced quote Report comment

votes Against and says:

At Eleuther AI, we’ve spent five years working to make large-scale AI research more transparent, accessible, and ethical, with a strong focus on data work.
AI Verified source (Oct 7, 2025)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Stefano Maffulli
Executive director of the Open Source Initiative

I am this person! Add a sourced quote Report comment

votes Against and says:

Open source AI requires: unrestricted access to trained parameters/weights; complete code used to train the system and prepare training data; full dataset (unless legally impossible).
AI Verified source (Oct 6, 2025)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Ana Klimovic
Assistant professor of computer science at ETH Zurich and Swiss AI Initiative steering-committee member

I am this person! Add a sourced quote Report comment

votes Against and says:

I’m really excited about the release of the fully open Apertus model and the follow-up work to come.
AI Verified source (Sep 17, 2025)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Antoine Bosselut
Professor at EPFL and co-lead of the Swiss AI Initiative

I am this person! Add a sourced quote Report comment

votes Against and says:

Apertus demonstrates that generative AI can be both powerful and open. [...] The release of Apertus is not a final step, rather it’s the beginning of a journey, a long-term commitment to open, trustworthy, and sovereign AI foundations, for the public...
more AI Verified source (Sep 2, 2025)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate

Loading more opinions...

More h/open-source votes

Open-source AI is more dangerous than closed-source AI

Democratic middle powers should jointly develop frontier AI and share its benefits globally

Back