Key Takeaways

  • OpenAI's advanced GPT 5.6 models—dubbed 'Soul,' 'Terra,' and 'Luna'—are only accessible to 20 pre-approved companies, a restriction directly influenced by US government directives after discussions with the Trump administration.
  • Model distillation, exemplified by Alibaba cloning Claude 'many, many times,' shows how quickly sophisticated AI can be copied and resold, making control over advanced capabilities nearly impossible once released.
  • John Coogan introduced the chilling 'six-month gap' analogy: advanced AI could generate novel bio-weapons or cyber threats much faster than humans can develop effective antidotes or defenses.
  • Despite cooperating, OpenAI cautions against government intervention becoming the norm, arguing it keeps vital tools from developers, enterprises, and "cyber defenders" who need them.

The Iron Curtain Descends on Advanced AI

Imagine a technology so potent, its release is governed by national security directives. That's precisely what's happening with OpenAI's newest, highly advanced GPT 5.6 models—internally known as 'Soul,' 'Terra,' and 'Luna.' John Coogan revealed these models are currently limited to just 20 pre-approved companies, a direct result of discussions with the Trump administration. While OpenAI acknowledged these discussions, they also pushed back, stating, “We don't believe this kind of government access process should become the long-term default.” This isn't just about protecting intellectual property; it's about controlling a capability deemed too dangerous for broad access.

This limited release signals a new era for AI development: one where the most cutting-edge models are effectively state secrets, tightly controlled and monitored. For ambitious founders, this means the 'open' in OpenAI might become a historical footnote. The frontier of AI innovation is now under the shadow of government oversight, forcing a re-evaluation of how you access, build with, and even perceive the future of this technology.

The 'Six-Month Gap': A New Asymmetric Risk

Perhaps the most unsettling insight from this conversation is Coogan's 'six-month gap' analogy. He posits a world where advanced AI could create a novel bioweapon or a devastating cyber-attack significantly faster than our best minds could engineer a countermeasure. “The problem that I'm worried about now is if there is a if there is a world where building the counter building the counter like the good version of something takes longer than six months,” Coogan explained. This isn't a theoretical concern; it highlights an urgent, asymmetric risk. The ability to generate threats is outstripping the ability to defend against them, with AI acting as the accelerator.

This gap isn't just about state-level threats. The rapid spread of model distillation—where companies like Alibaba have “distilled Claude many, many times” and are reselling tokens at a "staggering" rate, as Coogan notes—means that even restricted models can quickly escape control. The accelerating pace of open-source AI further compounds this. What takes state actors months to secure, could take rogue groups weeks to replicate and weaponize if the base models are out there. This puts founders in a precarious position: relying on these tools means accepting an increased, invisible layer of systemic risk.

What to Do With This

The era of unrestrained access to the most powerful AI is over, at least for now. Founders, stress-test your AI supply chain. Identify which foundational models power your products and research how government regulations or provider restrictions could impact access. Beyond that, build robust internal AI ethics and security protocols today, even if you're not training your own models. The 'six-month gap' means an unexpected, AI-driven bio or cyber event could fundamentally alter your market or even your operating environment. Proactively engage with policy discussions around AI safety—your future business depends on the guardrails that are (or aren't) being built.