Discord users reportedly accessed Anthropic Mythos Preview without approval

In short: A group of Discord users reportedly gained unauthorized access to Anthropic’s restricted Mythos Preview AI tool by using clues from leaked data and guessing where it was hosted.

What happened

Bloomberg reported that some Discord users found a way into Anthropic’s Mythos Preview, an AI system that is meant to be tightly controlled. Mythos has been described as a tool that can help find security weaknesses in software and networks.

According to the report, the group looked at information from a recent data breach involving Mercor, a company connected to AI training work. Using that material, they made what Bloomberg called an “educated guess” about where Mythos was located online, likely meaning a web address, similar to guessing the location of a locked room because you have seen how other doors in the same building are labeled.

The report also says one person used access rights they already had through work for an Anthropic contracting firm. As a result, the group allegedly reached not only Mythos Preview but also other unreleased Anthropic models.

Bloomberg said the group appeared to avoid drawing attention, and used the tool only for simple website projects rather than attacks.

Why it matters

This story highlights a basic problem with powerful digital tools. Even if a company tries to limit who can use them, a small mistake in how access is set up can be enough for outsiders to get in. For everyday users, the takeaway is that security often depends on simple controls, like who has the keys and whether doors are labeled in predictable ways.

Source: Wired