Positive users agree the warning on open-weight frontier models heightening cyber and bio risks is fair and seek mitigation ideas, while negative users dismiss the concerns as a fear cult driven by ego or morally wrong restrictions.
Based on 7 visible X reactions from 10 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@BlackHC Dario would be wrong. Public availability of capabilities is necessary to make society AI-resilient. It鈥檚 also morally revolting to have one or a few organizations (such as the current US regime) control frontier AI and decide who gets to use it for what.
@BlackHC We have already seen governments and large companies use LLM and AI in the most depraved way possible. Dropping bombs civilians, targeting people with medical issues in layoffs. The only risk is that big tech and governments loose the monopoly
@BlackHC They don鈥檛 pose risks. There is a fear cult that has captured the minds of very capable people. It has more to do with ego gratification for the members of the cult. 鈥淚鈥檓 responsible enough to be hands-on, but you aren鈥檛.
@BlackHC Who monitors and who becomes a risk? The biggest security risk on this planet at this moment is a rapist fascist president of the biggest economy of the world. Do you read the news?:)
@BlackHC that鈥檚 a fair point, how do you think we mitigate that?
Dario wouldn't be wrong. While open-source/weight models are great and incredibly useful, having peak-frontier open-weight models absolutely increases the risks they pose, incl the ones named by OP Note: the frontier is jagged and the concerns apply mainly to cyber, biorisk, surveillance, etc. Having an open-source frontier model that's e.g. mainly frontier at writing CUDA kernels wouldn't be that concerning at all
Positive users agree the warning on open-weight frontier models heightening cyber and bio risks is fair and seek mitigation ideas, while negative users dismiss the concerns as a fear cult driven by ego or morally wrong restrictions.
Based on 7 visible X reactions from 10 accounts; directional sample.
Ask a question below.
Published answers will appear here.
I don't disagree with you. The tweet was written in the context of K3, which seems to be close or at the frontier, at least acc to some reports. The point is that having peak frontier capabilities in bio and cyber creates additional risks. While closed models can be served using KYC and with monitoring, open-weight models can be run locally by anyone with little monitoring Did you miss https://www.nytimes.com/2026/07/10/us/politics/ai-terrorism-boko-haram-nigeria.html ? Even if we manage to prevent such abuses, safety layers can be abliterated in open models Do you need a public casualty event to be convinced that bio risks or cyber risks pose serious risks? Why should I trust you with models with such capabilities and you me?
Like any other technology, terrorists will use it to their advantage. Therefore, it is important that everyone has the ability to use this technology and modify it anyway they want, so that they can become resilient to various nefarious actors. It is ridiculously easy to designate anyone you don鈥檛 like as a terrorist. My confidence in whatever regulatory body to not do this is approximately zero. Personally, I think the law should force every model that is served to the public, including frontier models, to be open-sourced. I don鈥檛 believe this will happen, so I hope that economic pressure will make it happen.
@togelius So the only good thing that stops a bad guy with a gun is a random civilian with a gun?