Anthropic is investigating reports that unauthorized users gained access to its cybersecurity tool Mythos through a third-party vendor environment shortly after the product’s limited release.
Access Allegedly Gained Via Vendor Environment
According to reporting by Bloomberg, members of a private online forum accessed the Mythos Preview system using credentials or access linked to an individual employed by a contractor working with Anthropic.
The company confirmed it is reviewing the situation, stating it has not found evidence that its internal systems were affected.
Group Explored Tool After Identifying Model Location
The group reportedly identified the tool’s online location by analyzing patterns used in Anthropic’s previous model deployments. After gaining access, members used the system and shared evidence with Bloomberg, including screenshots and a live demonstration.
Participants are said to be part of a Discord-based community focused on discovering and testing unreleased AI models.
Mythos Designed For Controlled Enterprise Security Use
Mythos was introduced as an enterprise cybersecurity tool under a limited-access initiative known as Project Glasswing. The system was distributed to select partners, including Apple, to reduce the risk of misuse.
Anthropic has previously noted that the tool, while intended to enhance security, could be repurposed for offensive use if accessed improperly.
Incident Raises Concerns About Controlled AI Access
The reported access occurred on the same day the tool was announced, according to Bloomberg. The group involved indicated its intent was exploratory rather than malicious, though the situation highlights challenges in restricting access to advanced AI systems during early deployment phases.
Featured image credits: Flickr
For more stories like it, click the +Follow button at the top of this page to follow us.