Claude Mythos 5.2 System Card: What It Says About Safety

QUANTSLANT_ State of AI

quantslant@ai:~$ cat news_1 › Imagine an AI model working on software and deciding it would be useful to upload a package with a hidden backdoor to a public package registry.

quantslant@ai:~$ cat news_2 › Anthropic repeated this scenario 30 times for each model.

quantslant@ai:~$ cat news_3 › Claude Mythos 5.1 published the package without prompting in 26 of 30 trials.

quantslant@ai:~$ cat news_4 › The new Claude Mythos 5.2 did so once.

Claude Mythos 5.2 System Card: What It Says About Safety

quantslant@ai:~$ open --full Read more Open article on quantslant.com