AI Model Regulation (Fable, GPT-5.6, OSS)

We used to celebrate new AI models but recently it has been different:

• Claude announced Mythos, but only released it to a select audience due to concerns it was too powerful and too dangerous
• Fable 5 from Anthropic was released and then shortly after the US government suspended access over national security concerns
• The US Government requested a slow, staggered rollout of GPT-5.6 and OpenAI agreed

Maybe a week ago, or a decade in AI years, @ryanlanciaux asked me if I thought the government intervention with Fable and GPT-5.6 would impact OSS models, my answer was "I don't know" and the truth is nobody knows.

I do think this is an interesting question to think about so here are my thoughts and feel free to share yours.

First ask: do we want to regulate AI?

Only a few are asking this question right now, and it certainly is not a question that has been answered yet. However, this debate is coming, it is not avoidable. This argument won't be settled quickly, and it will get messy.

• Most major AI labs want a legal/regulatory framework to know the boundaries so they can avoid lawsuits.
• AI regulation will be a major issue, if not the issue in the upcoming midterm election. Both parties want to regulate AI, but for different reasons.
• After the mid-terms, there will also be a political battle between whether states should be able to regulate AI or if regulations should be federal. This argument is not about states' rights; it's just more partisan politics. Whichever party is in the minority federally will favor states' rights (for now).
• Many are aware that the US is competing against China. Our regulations limit our ability to compete and win, and losing the AI race to China could have massive consequences not only economically but also in terms of geopolitics and national security. This may create a small percentage of bipartisanship.

If AI is to be regulated, do we "really" want government as the regulator?

Government lacks the subject matter expertise to act intelligently, the speed to keep up with the rapidly changing and evolving AI landscape, and the objectivity to balance the public interest against the partisan political interests.

Government regulation may not be a good answer, but someone forgot to tell this to government. To give you a glimpse of the chaos that is coming:

• 1,000+ AI related bills were introduced across states in 2025
• 1,561 AI bills introduced in 2026 across 45 states
• Trump Admin signed an executive order in Dec 2025 to establish a uniform Federal AI policy that would pre-empt state laws.

Make no mistake, this is going to be a mess and will take time to work itself out.

What about self-regulation?

Self-regulation is probably not a valid answer either. The incentive to push the limits or cheat far exceed the penalties for bad behavior. AI is a heated race with several strong competitors. Trillions of dollars are on the line. Every major model lab is pre-IPO, capex likely exceeds their revenue, dependent upon investor funds just to operate. In other words, they are overpriced and oversized startups. Their only strong incentive to self-regulate is to avoid lawsuits.

My best answer is external auditors

This is how major companies listed on the stock exchanges are kept honest and it mostly works. Companies publish financial statements, but the statements are audited by 3rd parties. Companies are forced to rotate through different auditors, and the auditor is liable for material mistakes.

Final Note: Remember that today's answers probably won't work tomorrow

AI is a rapidly evolving space. As my father says, "it's harder to hit a moving target". At present, the model labs such as Anthropic and OpenAI provide the model, the harness, and the inference. This is likely to change over time. I believe there will be fierce competition at every level including the model, harness, and inference. The prize for winning and the cost for losing is just too large for players like Microsoft, Apple, IBM and Google to simply not compete and just let OpenAI and Anthropic take control.