Discord Users Gain Unauthorized Access to Anthropic's Mythos AI
Original: Discord Sleuths Gained Unauthorized Access to Anthropic’s Mythos
Why This Matters
Highlights security risks in AI model access controls and unauthorized use concerns
A group of Discord users gained unauthorized access to Anthropic's restricted Mythos Preview AI model through detective work involving a Mercor data breach and exploiting existing permissions from Anthropic contracting work.
Amateur sleuths on Discord obtained access to Anthropic's highly restricted Mythos Preview AI model, which is designed to find security vulnerabilities and has been carefully controlled due to its dangerous capabilities. The group used straightforward detective work, examining data from a breach of AI training startup Mercor and making educated guesses about the model's online location based on Anthropic's URL format patterns. One member leveraged existing permissions from working with an Anthropic contracting firm to access not only Mythos but other unreleased Anthropic models. According to Bloomberg, the group has only used the access to build simple websites to avoid detection rather than exploit its hacking capabilities.