"“My personal conclusion can however not end up with anything else than that the big hype ar”"
"“My personal conclusion can however not end up with anything else than that the big hype ar”"
"“My personal conclusion can however not end up with anything else than that the big hype ar”"
Claude Mythos was tested against curl, a widely used open source data transfer tool. Anthropic had claimed the model identified thousands of zero-days before launch, but the restricted program limits access to major organizations. Daniel Stenberg received a third-party report rather than running the analysis himself and did not have direct access to the model. The report said Mythos found five confirmed security vulnerabilities in curl’s 178,000 lines of code. A review determined three were already documented issues and one was a bug rather than a security hole. Only one low-severity vulnerability was confirmed and is scheduled for a late-June patch. Prior AI analyses found many more issues, while Stenberg suggested AI tools are better at finding security holes but Mythos may be less dangerous than claimed.
#ai-security-testing #claude-mythos #curl-vulnerabilities #open-source-software #software-supply-chain
Read at SecurityWeek
Unable to calculate read time
Collection
[
|
...
]