OpenAI is making headlines once again with the widespread public release of its latest advanced large language model (LLM), Sol. This powerful AI is reportedly on par with Anthropic’s Fable, a model whose capabilities once caused such a stir in the White House that its public access was briefly restricted. Given the previous government unease, a critical question emerges: how exactly did Sol get the green light for release?
The short answer, surprisingly, is that nobody seems to be entirely sure. The process for approving these cutting-edge AI models remains shrouded in mystery, leaving many experts and industry insiders scratching their heads. This lack of transparency raises significant questions about accountability and the standards being applied to safeguard the public.
The Murky Path to AI Approval
Even those close to the regulatory discussions admit to a lack of clear insight. Mina Narayanan, a senior research analyst at Georgetown’s Center for Security and Emerging Technology, candidly told TechCrunch that she doesn’t “have visibility into those exact processes.” This sentiment is echoed by Dean W. Ball, a former Trump policy advisor now with OpenAI, who noted in his newsletter that “nobody knows what the requirements are to get licensed.”
Andy Konwinski, a computer scientist and co-founder of Databricks, Perplexity, and the Laude Institute, underscores the systemic issue. He shared that he has yet to meet anyone, even within frontier AI labs, who fully comprehends the approval process. Konwinski describes this as an “existential problem,” highlighting concerns about who ultimately holds the power to make these crucial gatekeeping decisions.
Anthropic’s Fable offers a relevant precedent, having been briefly pulled from public access due to government concerns. While Anthropic stated they were in conversations with the government, developing classifiers to detect ‘jailbreak’ attempts and implementing ‘defense-in-depth’ strategies, the specific nature of these dialogues remains largely unknown. This pattern of vague assurances without concrete details appears to be a recurring theme.
Government Efforts and Industry Influence
Eighteen months into the Trump administration, despite an executive order last month outlining a roadmap for evaluating frontier models, specifics are still missing. The order, which followed weeks of internal debate, mainly clarified what won’t be happening. For instance, Sriram Krishnan, a former White House AI advisor, confirmed to the Financial Times that “there will not be an FDA for AI.”
There’s also no consensus on which types of models demand government scrutiny or which agencies should conduct these evaluations. For now, the Department of Commerce’s Center for AI Standards and Innovation (CAISI) appears to be taking the lead. However, the executive order mandates six cabinet agencies to finalize a process by early August, suggesting that the current approach is, at best, ad hoc.
OpenAI CEO Sam Altman mentioned on CNBC that their process involved discussions with officials like Secretary of Commerce Howard Lutnick and Secretary of the Treasury Scott Bessent. Yet, the identity of the technical experts who actually tested Sol and their evaluation methodology remains undisclosed. OpenAI declined to elaborate on the government’s specific process, instead pointing to external evaluations by organizations such as UK AISI, SecureBio, and Irregular, detailed in Sol’s safety card.
Like Anthropic did with Fable, OpenAI gave the government and selected users a preview of Sol before its wider release. However, the identities of these preview users and the criteria for their selection are unknown. Interestingly, OpenAI itself stated in a late June blog post that they “don’t believe this kind of government access process should become the long-term default,” indicating a desire for a different path forward.
The backdrop to these conversations includes reports of Altman reportedly offering up to 5% equity in OpenAI for the administration’s “Trump Accounts,” alongside OpenAI president Greg Brockman’s significant donations to Trump’s political operations. For external observers, it’s difficult to completely separate these financial and political activities from the government’s seemingly “lighter-touch” approach to regulating Sol.
Balancing Innovation and Oversight
From an industry standpoint, a hands-off regulatory approach might seem appealing, but one that relies heavily on personal connections with administration officials creates both uncertainty and potentially perverse incentives. Many experts, including Andy Konwinski, worry that true technical experts are not adequately involved in the model release process. He believes safety researchers, alignment researchers, interpretability researchers, and data professionals across the stack need a stronger voice.
Konwinski advocates for an “open commons” model, similar to the FDA or NIH, where researchers, government officials, and private companies collaborate to build consensus on safety. This contrasts with the current capitalistic incentives that push AI companies to recoup vast training costs by releasing models quickly. As Konwinski points out, companies have “very clear legal obligations and fiduciary responsibility” built into their operating procedures.
Dean W. Ball suggests a future where government-licensed, third-party auditing organizations evaluate frontier labs’ safety measures. Konwinski also sees potential in new institutional formats like focused research organizations (FROs) to allow more objective experts from academia and non-profits to access and assess advanced models. These proposals aim to bring greater rigor and impartiality to the process.
The Future of Frontier AI Governance
For now, the secrecy surrounding AI development shows no signs of dissipating. This lack of transparency, however, is likely to fuel political challenges for an industry that the American public increasingly views with skepticism. As University of Wisconsin-Madison computer science professor Remzi Arpaci-Dusseau observed, “There’s not a sense that responsible people are driving forward these changes.”
Echoing these concerns, David Siegel, the computer scientist who founded Two Sigma, shared a sobering vision at a recent conference: “Imagine a situation… [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 state of affairs, it seems we don’t have to imagine it at all.
Source: TechCrunch – AI
