Django co-creator Simon Willison criticizes Daytona for withholding sandbox source code, alleging AGPLv3 violations

@simonw @daytonaio I can reverse engineer it in a few weeks manually or with AI in days...

@simonw @daytonaio Ivan was on @latentspacepod a month ago talking about how being open source benefits their cloud product. What changed in a month?

@simonw @daytonaio yeah, I don't get this strategy:

1. scare your existing users with security concerns

2. enrage users that choose you for open source

@simonw @daytonaio idk man i think the landscape has changed dramatically post mythos, it no longer makes sense to assume open source will *increase* security

@simonw @daytonaio Have you seen the new microvm sandboxes by docker?

Now when any mythos gpt-5.5 cyber vulnerability is reported and fixed in open source code. Anyone can deploy an agent to monitor the repo 24x7 for PR's that look like exploit fixes and with agents be exploiting that vulnerability before the fix in flight can be pushed into prod. Seems kind of reasonable to be concerned if your providing secure sandboxes as a service. Github needs to be more nuanced with their visibility options.

This was possible before but the speed at which vulnerabilities are reported now and the ability to have smart bots sitting on repos looking for incoming exploit patches make this a security process issue that has to be considered.

@simonw @daytonaio If you can't show us the code, you've already answered the security question. "Trust us" is a product category that got disrupted years ago.

@simonw @daytonaio This doesn’t make any sense tho. So they’re going to fully change their code overnight? Because the current version is forever on the internet?

@simonw @daytonaio https://www.docker.com/blog/docker-sandboxes-run-claude-code-and-other-coding-agents-unsupervised-but-safely/

@simonw @daytonaio Could it be due to blatant copying and IP theft from OSS? Raise the bar and close source it?

@simonw @daytonaio "Trust us, our sandbox is secure" ranks up there with "the cheque is in the mail."

@simonw @daytonaio Security by obscurity. What a great idea...

@simonw @daytonaio the closed-source security product is a classic

@simonw @daytonaio I think there is a difference between sandbox security and repo exposure, but yeah the optics are pretty bad

@simonw @daytonaio http://cal.com did the same thing and I'm just not sure I buy it...

Game theory wise, if a company has PMF then they have more incentive than any hacker to spend on security, so should just pony up to a few AI enabled pentest services to scan everything over

@simonw @daytonaio 😎

@simonw @daytonaio I understand this more as a move after Fable/Mythos. If you can have access to the latest models you can use them to threat-mitigate. If you don’t, that means someone else does and you don’t. A scary place to be

@simonw @daytonaio It is a fair critique. Daytona moved their production codebase to closed source on June 20, citing the risk of AI-assisted vulnerability discovery.

Hard to argue that hiding code solves the underlying isolation problem, but it is a massive shift in their product strategy.

@simonw @daytonaio Are they? What if they’re saying that they rely on a technical moat that’s impossible to maintain in the age of AI? Just as likely?

@0xb33bs @simonw @daytonaio 👀