It was discovered that one of these unauthorized individuals exploited his privileges as a third-party contractor for Anthropic to gain access to Mythos, while others employed different tactics.
An insider informed the agency that multiple participants of a private online forum, Discord, obtained access to Mythos on the same day Anthropic initially revealed intentions to offer the model to a restricted number of companies for evaluation. Since that time, the group has consistently utilized Mythos to test new AI models and has not initiated cybersecurity assessments, the source mentioned.
Anthropic revealed its new language model, Claude Mythos Preview, on April 7. The company indicated that it can recognize and then take advantage of essential vulnerabilities in all leading operating systems and web browsers at the user's request. Numerous vulnerabilities have persisted for decades and have gone unnoticed despite ongoing human testing.
source: bloomberg.com
An insider informed the agency that multiple participants of a private online forum, Discord, obtained access to Mythos on the same day Anthropic initially revealed intentions to offer the model to a restricted number of companies for evaluation. Since that time, the group has consistently utilized Mythos to test new AI models and has not initiated cybersecurity assessments, the source mentioned.
Anthropic revealed its new language model, Claude Mythos Preview, on April 7. The company indicated that it can recognize and then take advantage of essential vulnerabilities in all leading operating systems and web browsers at the user's request. Numerous vulnerabilities have persisted for decades and have gone unnoticed despite ongoing human testing.
source: bloomberg.com
