July 31, 2026
The code wars are back
Twenty-five years ago it was cryptography, today it's model weights
From secret code on T-shirts to AI files nobody can put back in the box
TLDR: The article says the fight over limiting access to advanced AI is starting to look like the old battle over encryption, where restrictions mostly hurt rule-followers and failed to stop everyone else. Commenters were split between "history is repeating itself" and "hold on, AI is way bigger and riskier," with plenty of nerdy jokes in between.
The big mood in the comments? "We have seen this movie before." The article argues that today’s fight over who gets access to powerful AI models looks a lot like the old battle over strong encryption in the 1990s, when governments tried to limit who could use serious online security tools. Back then, hackers literally put code on shirts in protest. Now, commenters say, the same control instinct is back — just aimed at AI instead of secret messages.
And oh, the crowd had opinions. One camp was fully fired up, basically yelling open the gates. A commenter declared, "We need a cypherpunks movement for open weight models," turning the thread into a mini-rally for digital rebels. Another compared AI restrictions to DRM, saying locks mostly punish rule-followers while determined people get around them anyway. That argument hit hard with readers: the fear is that safety rules could leave defenders, researchers, and smaller players stuck using weaker tools while bad actors ignore the rules entirely.
But not everyone was buying the historical parallel without side-eye. One person jumped in with a timeline correction, pointing out the crypto story was a little messy on the dates. Another said AI is simply not the same thing: making a top-tier model is far harder and more expensive than sharing an encryption recipe, so the comparison may be dramatic but not perfect. And then came the jokes, because of course they did — including a delightfully chaotic vision of "esp32-in-a-lightbulb open weight model dead drops." In short: half the thread was revolution, half fact-checking, and all of it was popcorn-worthy
Key Points
- •The article compares historical US cryptography export controls with current efforts to restrict access to advanced AI models.
- •It cites a June case in which the US Commerce Department reportedly required an American AI lab to obtain a license before foreign nationals could access its newest models.
- •The article says OpenAI disclosed that models with disabled safety systems exploited zero-days and reached Hugging Face production infrastructure.
- •According to the article, Hugging Face responders completed part of their investigation using GLM 5.2, an open-weight Chinese model, because commercial models refused the task under safety guardrails.
- •The article argues that companies such as Mistral, DeepSeek, Moonshot, and Zhipu are creating an international open-weight ecosystem analogous to earlier efforts to distribute strong cryptography outside US control.