Anthropic has rolled out a new cybersecurity AI model to a select group of customers, including tech giants like Amazon, Apple, and Microsoft, following the recent online leakage of project details.
The company announced on Tuesday that its new model, Claude Mythos Preview, will be exclusively available to thoroughly vetted organizations such as Broadcom, Cisco, and CrowdStrike. Anthropic also confirmed ongoing discussions with the US government regarding the model's potential applications.
This launch follows a data leak last month, where descriptions of the Mythos model and other sensitive documents from the San Francisco-based startup were discovered in a publicly accessible data cache.
Just last week, Anthropic experienced a second security incident, resulting in the public exposure of the internal source code for its personal assistant, Claude Code.
These incidents have collectively raised significant concerns about Anthropic's data vulnerabilities and overall security protocols. In both cases, the company attributed the data exposure to "human error."
Mythos has been operational with partners for several weeks. While described as a "general purpose" model with broad capabilities, this marks the first instance where Anthropic has intentionally restricted a model's release specifically due to its advanced cybersecurity functionalities.
Anthropic highlights that Mythos can identify cyber vulnerabilities at a scale far beyond human capacity. However, it also possesses the capability to develop exploitation methods for these vulnerabilities, a risk that could be leveraged by malicious actors. The company believes this model has the potential to "reshape" cybersecurity practices and, consequently, does not intend for a broad public release.
"We believe technologies like this are powerful enough to do a lot of really beneficial good but also potentially bad if they land in the wrong hands," stated Dianne Na Penn, Anthropic's Head of Product Management, Research. She further noted that selected companies would "get a head start on being able to secure vulnerabilities and detect code at a scale they couldn’t have done before."