OpenAI Launches the Safety Fellowship With Weekly Stipends of Nearly 4 Thousand Dollars for External Researchers
OpenAI announced on April 6 the Safety Fellowship, a pilot program that pays external researchers to study the risks of advanced artificial intelligence. Each fellow receives a weekly stipend of $3,850, which works out to more than $200,000 annualized, plus approximately $15,000 per month in compute resources and direct mentorship from OpenAI researchers. The program runs from September 14, 2026 to February 5, 2027, and applications are open until May 3, with selected candidates expected to be notified by July 25.
The timing of the announcement, though, drew attention for a very specific reason. On the same day, The New Yorker published an investigation by Ronan Farrow revealing that the company had shut down three internal safety teams over the course of 22 months, and had also removed the word safely from its mission statement filed with the IRS.
On one hand, a new, well-funded program aimed at independent research on artificial intelligence safety. On the other, a string of internal decisions that went in the opposite direction. This tension is what makes the OpenAI Safety Fellowship more than just an opportunity for researchers — it raises real questions about how the company is handling safety in practice, not just on paper. 🤔
What the Safety Fellowship Is and How It Works
The OpenAI Safety Fellowship is described by the company itself as a pilot program to support independent research in AI safety and alignment and to develop the next generation of talent in the field. The pitch is straightforward: OpenAI funds researchers from outside the company so they can conduct AI safety studies without being on the payroll. That includes the freedom to define their own research scope within priority areas, access to API credits and compute resources, and competitive compensation for full-time dedication to the topic during the program period.
Fellows can work in person at the Constellation space in Berkeley or remotely. One important detail is that fellows will not have access to OpenAI’s internal systems. They receive API credits and compute power, but the program positions itself as research funding from a distance, not as a rebuild of the internal teams that were dissolved.
The initiative is also not restricted to artificial intelligence specialists. OpenAI is actively recruiting professionals in cybersecurity, social sciences, and human-computer interaction, in addition to computer science. Specific academic credentials are not required — the company stated it prioritizes research ability, technical judgment, and execution capability.
In theory, the model makes a lot of sense. The idea of bringing in outside voices to evaluate the risks of AI systems is an approach that has gained significant traction in the ethics and technology governance debate in recent years. Independent researchers tend to have fewer conflicts of interest than internal teams, and their findings typically carry more credibility with the scientific community and the general public. For those working in artificial intelligence safety, this kind of access to compute resources can make a huge difference in the quality and impact of the studies produced. 🧠
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
The Seven Priority Research Areas
The program defines a research agenda spanning seven priority areas. Fellows can direct their studies toward any of them, and by the end of the program in February 2027, each fellow must deliver a substantive output — whether a scientific paper, a benchmark, or a dataset. The areas are:
- Safety evaluation — methods for testing and measuring how safe AI models are across different use scenarios
- Ethics — investigation into the moral and social implications of developing and deploying advanced AI systems
- Robustness — research on how to make models more resistant to failures, adversarial attacks, and unexpected behaviors
- Scalable mitigations — development of safety solutions that can keep pace with model growth
- Privacy-preserving safety methods — techniques that allow evaluating and improving system safety without compromising sensitive user data
- Autonomous agent oversight — mechanisms for monitoring and controlling AI systems that operate independently, known as AI agents
- High-severity misuse domains — research focused on scenarios where AI misuse can cause significant harm
This agenda is broad and covers most of the topics the scientific community considers most urgent when it comes to AI safety. The inclusion of autonomous agent oversight is particularly relevant, given that the next frontier in artificial intelligence involves precisely systems that make decisions and take actions with little to no human intervention. 🔬
The Context That Complicates Everything
If the Safety Fellowship had been announced on any random day, it probably would have been met with near-universal enthusiasm. But April 6 was not just any day. Ronan Farrow’s investigation in The New Yorker, which came out on the same date, brought information that put OpenAI’s safety narrative under much more rigorous scrutiny than usual.
According to the report, the company dismantled three consecutive internal teams dedicated to AI system safety over the course of 22 months:
- The superalignment team was shut down in May 2024, following the departure of its co-leads Ilya Sutskever and Jan Leike. Leike wrote publicly upon leaving the company that safety culture and processes had taken a backseat to shiny products.
- The AGI Readiness team was dissolved in October 2024.
- The Mission Alignment team was disbanded in February 2026, after just 16 months of existence.
The report also captured a revealing moment. When a journalist asked to speak with OpenAI’s existential safety researchers, a company representative replied: What do you mean by existential safety? That is not, like, a thing. The statement, coming from within the very company that helped popularize the debate around existential AI risks, is surprising to say the least.
But it did not stop there. The removal of the word safely from the mission statement filed with the IRS is a detail that might seem small, but carries considerable symbolic and legal weight. Documents filed with government agencies like the IRS are not changed by accident or bureaucratic oversight — they reflect deliberate decisions about how an organization wants to officially position itself. When a company that built much of its reputation around a commitment to safety removes precisely that commitment from an official document, the question that naturally arises is: what changed? And more importantly, why? 🧐
This combination of factors created a scenario where the Safety Fellowship announcement ended up being read by many as an attempt to manage the narrative during a delicate moment. There is no way to say that for certain, of course, but the timing coincidence is hard to ignore. OpenAI is one of the most closely watched organizations in the world today, and every move it makes is analyzed with a fine-tooth comb by researchers, regulators, journalists, and the general public.
An External Fellowship Does Not Replace Internal Infrastructure
A point worth highlighting is that the program’s own format makes it clear that it is not a replacement for the internal teams that were dissolved. Fellows receive API credits and compute resources, but they do not have access to OpenAI’s internal systems. That means they cannot, for example, directly audit model training processes, evaluate proprietary data, or influence development decisions in real time.
In practice, the Safety Fellowship functions more like an academic research funding program than an operational safety structure. That does not diminish its value — external research is essential and generates important contributions. But it is very different from having dedicated teams inside the company, with full access to systems and the power to influence product decisions. These are complementary things, not substitutes, and the dissolution of internal teams is not offset by the existence of external fellows working remotely. 💡
Safety and Ethics in AI: Why This Matters So Much
The debate around the OpenAI Safety Fellowship is not an isolated one. It is part of a much larger conversation that the artificial intelligence industry is having with itself and with the world about what it means to develop technology responsibly. Ethics in AI has moved beyond being a niche academic subject and has become one of the central topics on the global technology agenda, with regulators in Europe, the United States, and several other regions trying to create frameworks that ensure AI systems are developed with some level of accountability and transparency.
In this landscape, the role of programs like the Safety Fellowship is genuinely important. Independent research on AI risks is scarce, underfunded, and often marginalized within major organizations in the industry. When a company the size of OpenAI decides to put money into this, it sends a signal to the entire market that this type of work has value. That can encourage other companies to do the same, expand the base of researchers dedicated to the topic, and in the long run contribute to artificial intelligence being developed with greater awareness of its impacts.
The problem, of course, is that signals need to be consistent to be credible. A company cannot simultaneously dismantle its internal safety teams, remove references to safety from official documents, and launch an external fellowship program with safety in the name, without that raising legitimate questions about its real priorities. Ethics is not a public relations exercise — it needs to be present in day-to-day operational decisions, in hiring, in organizational structures, and in strategic choices that rarely show up in press releases.
The Impact Beyond the AI Industry
What happens inside OpenAI does not stay inside OpenAI. Trust in the safety commitments declared by frontier AI companies functions as a market signal that affects capital allocation across the entire artificial intelligence infrastructure. Investors evaluating the AI infrastructure sector closely follow OpenAI’s spending trajectory and the credibility of its operational priorities — especially at a time when there is growing overlap between blockchain-based systems and AI infrastructure.
AI tokens, autonomous agent protocols, and decentralized physical infrastructure projects are increasingly connected to the artificial intelligence ecosystem. When the safety credibility of a company like OpenAI is questioned, the cascading effect can reach sectors far beyond language model development. The market pays attention not only to what companies announce, but to the consistency between the message and concrete actions.
What Researchers and the AI Community Are Watching
Within the artificial intelligence research community, the reaction to the Safety Fellowship has been mixed. Some researchers view the program with cautious optimism: after all, any additional funding for AI safety studies is welcome, regardless of who is footing the bill. Another group, however, raises a legitimate concern about the real independence of researchers funded by OpenAI and about what happens when research findings are not favorable to the company or contradict decisions it has already made internally.
This tension between funding and independence is not unique to OpenAI — it is a structural challenge that the entire field of technology research has faced for decades. But it takes on especially relevant dimensions when we are talking about artificial intelligence safety, a field where the risks are potentially enormous and where the credibility of scientific findings can have direct implications for public policy, regulations, and the choices that billions of users make every day when interacting with AI systems.
The methodological and editorial independence of Safety Fellowship participants will therefore be one of the most closely watched aspects as the program progresses. The first cohort of fellows will have their results published in early 2027, and it is from those deliverables — papers, benchmarks, and datasets — that the market and the scientific community will be able to assess whether the program produced genuinely independent research or served more as a stamp of legitimacy for the company. 🔍
What to Expect Going Forward
With the application deadline set for May 3 and notification of selected candidates expected by July 25, the coming months will reveal a lot about the profile of researchers OpenAI aims to attract and about the level of autonomy they will actually have in practice. The diversity of accepted fields — from cybersecurity to social sciences and human-computer interaction — suggests the company wants to broaden the AI safety conversation beyond traditional machine learning circles, which is a positive sign.
At the end of the day, the OpenAI Safety Fellowship is a program that deserves close attention, both for what it promises and for the questions it raises. Independent research on artificial intelligence risks is necessary, urgent, and has real potential to make a difference. What remains to be demonstrated is whether OpenAI is willing to sustain this commitment beyond the announcement — in the tough decisions that no official press release will cover, but that will define, in practice, what the company truly thinks about ethics and AI safety. Whether the fellows who receive these grants can actually influence model development even without internal access is the question the first cohort will begin to answer in early 2027. 🚀
