Anthropic investigating possible breach of its Mythos AI model
Anthropic, the artificial intelligence company behind the chatbot Claude, is currently investigating a potential breach of its new AI model, Mythos. Mythos was recently rolled out to a select group of companies to assist in identifying software vulnerabilities. According to an Anthropic spokesperson, there has been a report of unauthorized access to Mythos from a third-party vendor environment.
Anthropic collaborates with a small number of third-party vendors to develop its AI models. While the company has not identified any breaches outside of its vendor environment or compromises to its systems, they are taking the reported breach seriously. The investigation into the alleged breach of Mythos was confirmed by Anthropic following a report by Bloomberg, which stated that unauthorized users had gained access to the tool.
Mythos was released as part of Project Glasswing, where it was presented as a more effective AI system for detecting software vulnerabilities compared to other existing models. Initially, only major companies like Amazon, Apple, Cisco, JPMorgan Chase, and Nvidia were granted access to Mythos. This limited distribution was a precautionary measure to prevent potential exploitation of the AI model by malicious actors.
Concerns have been raised by federal officials, security experts, and leaders of global institutions like the International Monetary Fund regarding the implications if Mythos were to fall into the wrong hands. While Project Glasswing aims to help companies enhance their cybersecurity defenses, there is apprehension that Mythos could be utilized to target IT infrastructure in critical sectors such as banking, healthcare, and government agencies.
Alissa Valentina Knight, CEO of cybersecurity AI company Assail, expressed the urgency for organizations to prepare for the evolving threat landscape. She highlighted the speed and capabilities of AI in cyber attacks, emphasizing the need for proactive measures to counter potential threats.
