Some Unknown Group Is Reportedly Utilizing Claude Mythos With out Permission

In a really cagily-written story from Bloomberg, Anthropic confirmed Tuesday that it has obtained a report that an unauthorized thriller group is accessing Claude Mythos—the mannequin it says is too dangerous to release. “We’re investigating a report claiming unauthorized entry to Claude Mythos Preview by way of considered one of our third-party vendor environments,” says an Anthropic spokesperson’s assertion to Bloomberg.

Bloomberg apparently confirmed the obvious breach by taking a look at a stay demo and screenshots despatched over by a member of the group liable for the unauthorized entry.

In understandably obfuscatory language, Bloomberg explains that an nameless supply says they’re a member of an unnamed group that has abused their entry “as a employee at a third-party contractor for Anthropic” and employed “generally used web sleuthing instruments typically employed by cybersecurity researchers,” to achieve some type of entry to the mannequin.

However don’t fear, this secret group that apparently has entry to essentially the most feared piece of know-how on the earth is “excited by enjoying round with new fashions, not wreaking havoc with them,” the supply apparently defined to Bloomberg.

The sequence of occasions within the obvious breach seems one thing like this:

• A Discord group exists which makes use of bots to smell round on GitHub for details about unreleased AI fashions
• There was an information breach on the AI coaching startup Mercor
• The group mixed info from the Mercor breach with entry obtainable to Bloomberg’s supply as a result of they work for an Anthropic contractor
• This allowed the group to guess the net location of Claude Mythos
• The group has been freely messing round with Claude Mythos ever since April 7, the identical day because the announcement of Project Glasswing

So to recap: Anthropic says it has the scariest AI mannequin on the earth, and for what it’s value, a whole lot of powerful institutions seem to believe it. If we take Anthropic at its phrase, we’re all trusting it to not abuse this energy that it and solely it controls. Nonetheless, some unknown entity has accessed this scary AI mannequin, but when we take them at their phrase, they simply used it for some vibe coding exams and so they swear they’re not doing something evil with it.