The global race to build Artificial General Intelligence (AGI) is moving faster than ever. But as the pace accelerates, a critical question looms: can speed and safety truly coexist in the development of powerful AI systems?
A Spark Ignites: Safety Concerns Go Public
The latest flashpoint came from Boaz Barak, a Harvard professor currently working on AI safety at OpenAI. He publicly criticized the launch of xAI’s Grok model, calling it “completely irresponsible.” His issue wasn’t with the model’s capabilities, but rather its lack of transparency—no public safety evaluations, no system card, no technical documentation. In an industry trying to normalize basic safety disclosures, this omission was alarming.
Inside OpenAI: Transparency vs. Secrecy
Just weeks after Barak’s comments, former OpenAI engineer Calvin French-Owen offered a candid view from the inside. In a public post, he confirmed that OpenAI invests heavily in safety—tackling issues like hate speech, bioterrorism risks, and mental health triggers. But he also highlighted a critical flaw: much of this work is never published. Despite good intentions, the pressure to stay ahead appears to be overshadowing the need for openness.
The Speed-Safety Paradox
The situation reveals a deeper conflict faced by every leading AI lab: the Speed-Safety Paradox. On one hand, the demand to innovate quickly has never been higher. On the other, there’s a moral imperative to develop technology responsibly. Balancing these two forces is proving incredibly difficult.
French-Owen described OpenAI as operating in “controlled chaos” after tripling its team to over 3,000 people in a year. Rapid growth, he explained, often results in broken systems and reactive management. The competitive pressure to outperform rivals like Google and Anthropic only amplifies this effect.
A Culture Built on Urgency
One striking example is how OpenAI developed Codex, the AI coding assistant. French-Owen revealed the project was completed in just seven weeks. The pace was so intense that developers routinely worked past midnight and through weekends. While this breakneck development helped push boundaries, it also raised questions about the cost—both to the team and to long-term safety.
Why AI Labs Default to Speed
This isn’t simply a matter of negligence. The AI field is shaped by several powerful forces:
Competitive urgency: Being first to market offers massive advantages.
Cultural norms: Many AI labs evolved from hacker-style research groups that prioritize breakthroughs over protocol.
Invisibility of prevention: It’s easy to measure how fast a model trains. It’s much harder to measure the value of a disaster that was avoided.
All of this makes performance more visible than prudence—and companies often prioritize what they can prove.
Rethinking AI Development: A Call for New Standards
If the industry is to align speed with safety, it needs a fundamental shift in how success is defined. Here are a few starting points:
Safety by default: Publishing a comprehensive safety analysis should be as essential as launching a new model.
Shared industry rules: No company should be penalized for being careful. Transparency and safety must be industry-wide standards, not competitive disadvantages.
Responsibility beyond safety teams: Every engineer, not just those in risk departments, should feel personally accountable for what they’re building.
The True Measure of Success
The race to AGI is not simply about reaching the destination—it’s about how we get there. Speed alone is not victory. The real winner will be the organization that proves it’s possible to combine ambition with integrity, progress with precaution.
In the end, the world is watching not just for who builds the most advanced AI—but for who does it right.
