In a very cautiously written story from BloombergAnthropic confirmed Tuesday that it received a report that a mysterious unauthorized group is accessing Claude Mythos, the model that says it’s too dangerous to release. “We are investigating a report alleging unauthorized access to Claude Mythos Preview through one of our third-party vendor environments,” an Anthropic spokesperson’s statement to Bloomberg reads.
Bloomberg apparently confirmed the apparent breach by looking at a live demo and screenshots sent by a member of the group responsible for the unauthorized access.
In understandably obfuscating language, Bloomberg explains that an anonymous source says they are members of an anonymous group that has abused their access “as a third-party contractor worker for Anthropic” and has employed “commonly used Internet research tools often employed by cybersecurity researchers” to gain some form of access to the model.
But don’t worry, this secret group that apparently has access to the world’s most feared technology is “interested in playing with new models, not wreaking havoc on them,” the source apparently explained to Bloomberg.
The sequence of events in the apparent breach looks like this:
- There is a Discord group that uses bots to snoop around GitHub for information on unreleased AI models.
- Data breach occurred at AI training startup Mercor
- The group combined information from the Mercor breach with access available to Bloomberg’s feed because they work for an Anthropic contractor.
- This allowed the group to guess Claude Mythos’ online location.
- The group has been playing freely with Claude Mythos since April 7, the same day of the Project Glasswing announcement
Long story short: Anthropic says it has the scariest AI model in the world, and for what it’s worth, a lot of powerful institutions seem to believe it. If we take Anthropic at his word, we all trust him not to abuse this power that he and he alone controls. However, some unknown entity has accessed this terrifying AI model, but if we take them At their word, they just used it for some vibration coding tests and swear they’re not doing anything wrong with it.
