A critique from an OpenAI researcher has given us a glimpse into the ongoing internal struggle within the AI industry. This struggle is not against external competitors, but rather, a battle against itself.
The controversy began with a stern warning from Boaz Barak, a Harvard professor currently on sabbatical and working on safety measures at OpenAI. He labelled the release of xAI’s Grok model as “utterly reckless,” not for its sensationalized exploits, but for its lack of transparency: a public system card, thorough safety evaluations, and other fundamental elements of transparency that the industry has come to tentatively embrace.
Barak’s admonishment was both clear and necessary. However, a candid reflection by former OpenAI engineer Calvin French-Owen, shared just three weeks after his departure from the company, reveals a more complex narrative.
French-Owen’s account indicates that a significant number of OpenAI’s staff indeed focus on safety, addressing very real threats like hate speech, bio-weapons, and self-harm. However, his critique includes the revelation that “most of the work which is done isn’t published,” and he asserts that OpenAI “needs to do more to publicize it.”
This disclosure complicates the simplistic narrative of a responsible entity chastising an irresponsible one. Instead, it exposes a broader industry-wide conundrum. The AI industry finds itself ensnared in the ‘Safety-Velocity Paradox,’ a profound structural conflict between the demand to innovate rapidly to stay competitive and the ethical obligation to proceed with caution to ensure safety.
French-Owen paints a picture of OpenAI as a company in a state of managed turmoil, having tripled its workforce to over 3,000 within a year, where “everything breaks when you scale that quickly.” This frenetic energy is driven by the intense competition of a “three-horse race” to AGI against Google and Anthropic. The outcome is a culture marked by extraordinary speed, but also one shrouded in secrecy.
Take, for example, the development of Codex, OpenAI’s coding agent. French-Owen describes the project as a “mad-dash sprint,” where a small team constructed a groundbreaking product from the ground up in just seven weeks. This is the human toll of such velocity. In such a fast-paced environment, the slow, meticulous work of publishing AI safety research seems like a diversion from the race.
This paradox is not borne out of ill-intent, but rather a combination of powerful, intertwined forces. The overt competitive pressure to be the first, the cultural DNA of these labs that prioritize disruptive breakthroughs over methodical processes, and the simple issue of measurement: it’s straightforward to quantify speed and performance, but incredibly challenging to quantify a disaster that was successfully averted.
In today’s boardrooms, the tangible metrics of speed will almost always drown out the silent victories of safety. However, progress cannot be about blame; it must be about altering the fundamental rules of the game.
We must redefine what it means to launch a product, making the publication of a safety case as essential as the code itself. We need industry-wide standards that protect any company from being competitively disadvantaged for its diligence, transforming safety from an optional feature into a shared, non-negotiable foundation.
Above all, we need to foster a culture within AI labs where every engineer – not just the safety department – feels a sense of responsibility.
The race to create AGI is not about who reaches the finish line first; it’s about how we get there. The true victor will not be the company that is merely the fastest, but the one that demonstrates to the world that ambition and responsibility can, and must, coexist.
(Photo by Olu Olamigoke Jr.)
For related news, see: Military AI contracts awarded to Anthropic, OpenAI, Google, and xAI.
Interested in learning more about AI and big data from industry leaders? Attend the AI & Big Data Expo taking place in Amsterdam, California, and London. This comprehensive event will run alongside other leading events including the Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.
Discover more upcoming enterprise technology events and webinars powered by TechForge here. The question remains: Can speed and safety truly coexist in the AI race?