By John Q. Hosedrinker 
OpenAI CEO Sam Altman stated that the company does not control the Pentagon’s operational decisions regarding their AI products, even as the military reportedly uses AI in operations like the seizure of Nicolás Maduro and targeting in the conflict with Iran. This comes amidst employee and public concern that OpenAI has crossed ethical lines that rival Anthropic refused to, particularly after the Pentagon declared Anthropic a “supply-chain risk” for refusing a deal. Despite Altman’s assurances of legal use and efforts at damage control, Anthropic’s CEO accused OpenAI of “safety theater” and political motivations behind their Pentagon agreement.
Read the original article here
It seems there’s a rather significant admission coming from Sam Altman regarding OpenAI’s capacity to control how the Pentagon utilizes their artificial intelligence technology. The core of this revelation is that, despite OpenAI’s stated commitment to ethical development, they ultimately lack any real leverage over the Department of Defense’s application of AI, particularly when it comes to potentially invasive or destructive purposes like mass surveillance or autonomous weapons systems.
This admission essentially cuts through a lot of the initial optimism and promises that surrounded AI development, especially concerning the idea that these powerful tools could be guided towards solely beneficial outcomes for humanity. When the very creators of these advanced AI systems acknowledge they can’t dictate their usage by major governmental bodies, it raises serious questions about accountability and the true nature of control in this rapidly evolving technological landscape.
The idea that OpenAI, or any company for that matter, would be unable to prevent their technology from being used for mass surveillance or “kill bots” is quite concerning. It suggests a fundamental disconnect between the development of powerful AI and the ethical frameworks meant to govern its deployment. One might assume that a company focused on ethical AI would have built-in safeguards or mechanisms to prevent such egregious uses, but Altman’s statement implies these are either insufficient or simply non-existent in practice when dealing with powerful governmental entities.
It’s difficult to reconcile the pronouncements of being “more ethical” with the reality of not being able to control the military’s use of AI for potentially devastating purposes. This disparity leads to a cynical interpretation: that the pursuit of funding and government contracts might outweigh genuine ethical concerns. The financial incentives of these partnerships could, inadvertently or intentionally, lead to a hands-off approach regarding how the technology is ultimately wielded, even if it leads to morally questionable applications.
The notion of a company developing a product it cannot control is inherently problematic. It’s akin to building a powerful tool and then simply hoping that whoever acquires it will use it responsibly, without any real ability to enforce that hope. This lack of control becomes particularly alarming when the potential applications involve military action, surveillance, or anything that could impact civilian lives and freedoms on a grand scale.
Furthermore, the comparison to situations where a nation might hand over something as destructive as a nuclear weapon to an unstable actor comes to mind. While not a direct analogy, it highlights the inherent danger of relinquishing control over powerful technologies without robust guarantees of responsible use. The potential for unintended consequences, or even deliberate misuse, becomes amplified when that control is vested elsewhere, especially in entities with vastly different priorities and objectives.
This admission also brings up the question of why the Pentagon wouldn’t simply develop its own AI systems. With substantial funding available, they possess the resources to build bespoke hardware and train AI exclusively on military data, potentially avoiding the ethical quandaries associated with pre-existing AI models that might have been trained on a broader, and perhaps more problematic, dataset. Training AI internally could offer greater control and alignment with specific military objectives, rather than relying on external providers whose ultimate control is questionable.
The reliance on API keys and terms of service agreements, while standard practice, also appears to be an insufficient guardrail when dealing with governmental entities. History suggests that governments, particularly when national security is invoked, often find ways to circumvent or disregard such agreements when it suits their objectives. The “strong-arming” of the private sector by governmental administrations further exacerbates this concern, suggesting that adherence to terms of service might be more of a suggestion than a strict mandate.
The expressed desire to move towards alternative AI platforms like Claude, while understandable given these concerns, also reveals a broader market dynamic. It seems that even competing AI developers face similar challenges in controlling the application of their technology by governments. The assertion that “Anthropic can’t control this either” implies that this is not an isolated issue with OpenAI, but a systemic challenge within the AI industry when interacting with governmental bodies.
The progression of OpenAI’s stated mission, from advocating for open AI for ethical use to a more closed-source model for safety, and then to a for-profit structure seeking substantial funding, raises flags. Each step appears to involve compromises, culminating in the current situation where they acknowledge a lack of control over the Pentagon’s AI usage, despite past assurances of ethical deployment. This trajectory can be seen as a series of concessions, driven by financial needs and the desire to remain at the forefront of AI development.
Ultimately, the core issue boils down to a profound question of responsibility and foresight. While the potential benefits of AI are immense, the acknowledgement that creators cannot control its application by powerful military forces paints a concerning picture. It’s a scenario that mirrors cautionary tales from science fiction, where technological advancement outpaces our ability to manage its ethical implications, leading to outcomes that are far from the utopian visions initially presented. The admission suggests that, for now, the fate of how military AI is used rests largely on the discretion of those wielding the power, with little to no ultimate oversight from the entities that created the technology itself.