Prabowo Widodo
OpenAI is rolling out its latest advanced large language model (LLM), Sol, for wide public access, a move that immediately casts a spotlight on the perplexing and largely undefined regulatory landscape governing the deployment of frontier artificial intelligence. Sol is considered to be at least on par with Anthropic’s Fable, another powerful model whose capabilities and even ownership structure reportedly stressed out the White House enough to trigger its brief ban from public access. The stark contrast between the government’s swift, albeit temporary, intervention with Fable and the seemingly smoother path for Sol raises critical questions about the criteria, transparency, and consistency of AI regulation at a pivotal moment for the technology.
The Unclear Path to Public Release
The central enigma surrounding the release of these highly capable AI models remains: how exactly do they get the green light for public deployment? The short answer, according to numerous experts and even insiders, is that nobody is entirely sure. This lack of clarity fuels growing apprehension among researchers, policymakers, and the public alike, especially given the rapid advancements in AI and its potential societal implications.
Mina Narayanan, a senior research analyst at Georgetown’s Center for Security and Open Technology, expressed the widespread uncertainty in a statement to TechCrunch. "Frankly, I don’t have visibility into those exact processes, so yes, I don’t feel like I have enough information to say whether they’re adequate or not," Narayanan noted. While Anthropic had indicated engagement with the government, developing classifiers for "jailbreak" attempts and implementing defensive strategies, the specifics of this dialogue—and similar conversations with OpenAI—remain shrouded in mystery. This opaqueness makes it nearly impossible for external parties to assess the rigor or effectiveness of the approval mechanisms.
The sentiment is echoed within the industry itself. Dean W. Ball, a former Trump policy advisor who now works for OpenAI, candidly wrote in his newsletter last month that "nobody knows what the requirements are to get licensed." This stark admission from an individual with direct ties to both government policy and a leading AI developer underscores the profound absence of a standardized framework. Andy Konwinski, a prominent computer scientist and co-founder of Databricks, Perplexity, and the Laude Institute, further corroborated this, stating he has never encountered anyone, including employees at frontier labs, who fully comprehends the approval process. Konwinski articulated the profound implications: "It’s existentially a problem. Safety or not, it’s about who has the power to make decisions—who gatekeeps and decides on permissions?" His remarks highlight a core concern about accountability and the concentration of regulatory power without transparent checks and balances.
Government’s Ad-Hoc Approach and Policy Vacuum
Eighteen months into the Trump administration, the roadmap for regulating advanced AI models remains largely incomplete. Despite—or, as some critics allege, because of—the involvement of prominent industry figures in setting policy, little concrete clarity has emerged. Last month, after weeks of reported infighting within the administration, an executive order was finally published. This order ostensibly laid out a framework for evaluating frontier models, but its specifics largely remained undefined, beyond what would not be established. Sriram Krishnan, a former Andreesen Horowitz partner and senior advisor for AI in the White House until recently, definitively stated to the Financial Times, "There will not be an FDA for AI." This declaration dispelled hopes for a dedicated, comprehensive regulatory body akin to those overseeing pharmaceuticals or food safety, leaving a significant gap in oversight.
A critical point of contention continues to be the lack of consensus on which types of AI models warrant government scrutiny, and precisely which agency or agencies are equipped and mandated to perform these evaluations. While the Department of Commerce’s Center for AI Standards and Innovation appears to be taking an interim lead, the executive order broadly instructs six cabinet agencies to collaborate and determine a final process by early August. In the interim, what has emerged is, at best, a patchwork, ad-hoc system that lacks consistency and predictability.
OpenAI’s Engagement and External Scrutiny
OpenAI CEO Sam Altman acknowledged the engagement with government officials during a CNBC appearance, naming Secretary of Commerce Howard Lutnick, Secretary of the Treasury Scott Bessent, and US national cyber director Sean Cairncross as key interlocutors. However, the details surrounding these conversations—specifically, who the technical experts were that tested the models, and the methodologies employed for such evaluations—remain undisclosed. OpenAI, when approached by TechCrunch for further details on the government’s process, declined to elaborate. Instead, the company pointed to the results of several external evaluations conducted by organizations such as UK AISI, SecureBio, and Irregular, which are detailed in Sol’s safety card. While such third-party assessments offer some degree of validation, they do not fully address the transparency concerns surrounding the government’s internal approval mechanisms.
Similar to Anthropic’s Fable rollout, OpenAI granted government entities and a select group of users early access to Sol before its wider public release. However, the identities of these users and the criteria for their selection remain confidential. In a blog post published in late June, OpenAI did express a desire for a more structured future: "we don’t believe this kind of government access process should become the long-term default," indicating a willingness to collaborate with the government on developing a more robust and transparent path forward. This statement, while positive, underscores the current dissatisfaction even from within the industry regarding the existing informal arrangements.
The Shadow of Political Influence
The backdrop against which these high-stakes discussions and model releases are occurring includes significant, and potentially influential, connections between OpenAI leadership and the current administration. Reports indicate that Sam Altman reportedly offered as much as 5% of OpenAI’s equity for the administration’s so-called "Trump Accounts." Furthermore, OpenAI president Greg Brockman has been identified as the largest publicly-known donor to the Trump administration’s mid-term political operation. These financial and political ties make it challenging for outside observers to definitively separate such activities from the government’s seemingly more lenient or "lighter-touch" approach to regulating Sol, especially when contrasted with the earlier scrutiny faced by Anthropic’s Fable.
Anthropic’s Fable, for instance, experienced a brief but significant setback when it was pulled from wider public access. The US government reportedly forbade its use by foreign nationals, driven by a dual set of concerns. Firstly, there were genuine anxieties about users potentially "jailbreaking" the model to exploit its hacking capabilities, posing national security risks. Secondly, reports suggested personality clashes between Anthropic and the Trump administration contributed to the heightened scrutiny. The threat of an export ban—a powerful tool for government leverage—may have also motivated OpenAI to be particularly cooperative with the government’s (still largely unknown) requests, potentially shaping the regulatory trajectory for Sol.
Industry Perspectives: Calls for Clarity and Expert Involvement
While a hands-off approach to regulation might initially appeal to the industry, one that heavily relies on personal connections to administration officials creates an environment rife with uncertainty and potentially perverse incentives. Such a system undermines fairness, predictability, and ultimately, public trust.
Andy Konwinski voiced a significant concern that true experts in AI technology—including "safety researchers, alignment researchers, interpretability researchers, but also data people, and people from all over the stack"—are not playing a sufficiently central role in the model release process. He argues that their deep technical understanding is crucial for adequately assessing risks and ensuring responsible deployment. Konwinski champions an "open commons" approach as the optimal way to balance safety with innovation. He points to established models like the FDA, the NIH, or national laboratories, which successfully convene diverse stakeholders—researchers, government officials, and private companies—to forge consensus on complex safety issues. This collaborative model, he suggests, could provide a more robust and transparent framework for AI governance.
The inherent incentives of capitalism also play a significant role, having driven AI researchers for over a decade and even surfacing in legal battles, such as Elon Musk’s lawsuit challenging OpenAI’s corporate structure. Dean W. Ball highlights that the nature of the AI business necessitates companies to rapidly recoup substantial training costs shortly after their models are released, especially to maintain a competitive edge. This intense commercial pressure can inadvertently create a drive for faster deployment, potentially at the expense of comprehensive safety evaluations or adherence to nascent regulatory guidelines. Konwinski further points out the legal and fiduciary responsibilities built into operating procedures, noting that "even if their intentions are good, there’s very clear legal obligations and fiduciary responsibility that are built right into the operating procedures." This implies that even well-intentioned AI developers are bound by economic imperatives that may prioritize speed to market.
Proposed Solutions and Future Outlook
Looking ahead, there are proposals for establishing a more structured and transparent regulatory environment. Ball, in his aforementioned post, advocated for a future where third-party auditing organizations, officially licensed and accredited by the government, would be responsible for evaluating frontier labs’ approaches to safety. This model would introduce an independent layer of scrutiny and expertise. Konwinski, similarly, expresses optimism about new institutional formats, such as focused research organizations (FROs), which could empower more disinterested experts from academia and the non-profit sector to access and critically evaluate frontier models, thereby broadening the base of informed assessment.
For now, the pervasive secrecy surrounding the development and deployment of advanced AI shows no signs of abating. However, this lack of transparency is poised to generate significant political challenges for an industry that Americans are increasingly viewing with skepticism, as evidenced by recent surveys. Remzi Arpaci-Dusseau, a computer science professor at the University of Wisconsin-Madison, articulated this growing public unease at the Open Frontier conference last week: "There’s not a sense that responsible people are driving forward these changes."
At the very same event, David Siegel, the computer scientist who founded Two Sigma, one of the most successful quantitative hedge funds, issued a stark warning to attendees. He urged them to "imagine a situation, which I think would be very bad, [where] a small number of firms control the technology; the government, in their secretive laboratories, is evaluating whether or not the technology is suitable for use; and the general public and scientific community doesn’t really have any access to any of that stuff." Given the current trajectory, with powerful AI models being released under opaque circumstances, and regulatory frameworks lagging behind technological advancements, it appears that this dystopian scenario may not require much imagination at all. The imperative for clear, transparent, and expert-driven governance of advanced AI has never been more urgent.
