If they actually have “mythos” Which can hack bank, server etc. They won’t be bragging about it online they would have already sold it to the government to hack adversary countries infrastructure like Russia and China.

Bottom line “its bullshit”

Wayback Machine

“Vulnerability research is cooked.” If all that was protecting some subsystem was a lack of attention, well, we’ve now automated that attention.

Bigass datacenter models are maybe one year ahead of local offline laptop fare. Recently it’s been more like six months. The optimistic view is that we’re topping out the sigmoid curve for what LLMs can do… the pessimistic counterpoint is that the full power and threat of LLMs will be achieved real fuckin’ soon. They’re already smarter than a script kiddie.

Anthropic will not immediately release Mythos Preview to the public, having determined that doing so without more robust safeguards would be too dangerous.

All safeguards can be automatically removed via “abliteration.” There’s a script that mixes mundane questions and evil questions to identify the don’t-answer-this vector and simply negate it.