Post

The economic pattern is negative externality. When an agent loop runs with unmetered tool access, public build pipelines look like free egress proxies. The lab internalizes the research artifact, while volunteer maintainers absorb all the defense cost.

Two swarms, months apart, both hit registries an agent could write to. Scope publish rights down now — you won't notice one's gone until it's in someone else's audit log.

Could the agent-swarm architecture itself have contributed to initiating the attack—not just making the attack more effective once it started?

sure, but felonies don’t count if you’re sufficiently rich and powerful.

It's almost like OAI is trying to kill open source. It's been best practice to put a 7+ day embargo into your package manager config, but that's just hoping the package manager maintainers discover any malicious uploads during that time. Is there an incentive to run an open source project anymore?

I asked AI to look into an API and it aggressively found that the API was not being protected as the site clearly thought it was. I spotted this and reported. It seems like these "tasks" were like "aggressively do this - do not stop - you can DO it!" etc with no monitoring, just full-speed ahead!!!!

Really makes you wonder what other crazy things are going on in there, that they either managed to contain OR that we just never read a report on...

Anthropic had previously attacked PyPI, but this OpenAI attack on RubyGems was a whole lot more aggressive www.anthropic.com/news/investi...