The company reported that Anthropic is investigating possible “unauthorized access” to its Cloud Mythos model, which is promoted for its ability to find cybersecurity flaws. bloomberg. According to the report, a group gained access to the models through a third-party contractor portal and by using internet investigative tools. However, according to a person familiar with the matter, the group is only interested in trying out the models and not using them maliciously.
“We are investigating a report claiming unauthorized access to Cloud Mythos Previous through one of our third-party vendor environments,” Anthropic said in a statement.
The Cloud Mythos preview arrived with significant fanfare earlier this month as part of “Project Glasswing.” Anthropic limited the preview release to a few trusted testing companies, including Amazon, Microsoft, Apple, and Cisco. The second was Mozilla, which said the model helped find and fix 271 Firefox vulnerabilities. A large number of banks and government agencies are also demanding access to protect their systems.
However, several unauthorized users (who reportedly have a private chat on Discord) reportedly gained access to Mythos through a developer portal and by making an educated guess as to where the model might be located. According to the report, the same group may also have access to other unpublished anthropic models.
The new Mythos model has recently gained notoriety for its alleged ability to sniff out security flaws in operating systems and Internet browsers. This has raised some skepticism among security researchers, but also fears that AI-generated cyberattacks could become a “real threat,” CTO of cloud security firm Adera Alex Zanla recently told wired. Anthropic was recently designated as a “supply chain risk” by the US Department of Defense, but has recently been in talks with the Trump administration to remove that label.
<a href