IIRC from the Anthropic report, the alleged danger of Mythos isn’t that it finds more vulnerabilities than previous models, but that it’s significantly more successful at exploiting them. Which this doesn’t seem to test.