New serious vulnerabilities spiked around release of Claude Mythos Preview

17 pointsposted 3 hours ago
by cubefox

4 Comments

hoppp

an hour ago

How are these reports verified to be valid? If there are too many some could be hallucinations too.

nextaccountic

43 minutes ago

The best case scenario for AI companies is, people receive those bug reports, look at the model that produced it and not even look at the details, just apply the fix mindlessly

This gives Anthropic a staggering amount of power. Oh it came from Mythos? We will just lose time trying to analyze it, better apply the fix ASAP

stingraycharles

39 minutes ago

> The best case scenario for AI companies is, people receive those bug reports, look at the model that produced it and not even look at the details, just apply the fix mindlessly

Do people maintaining serious software do this, though?

solenoid0937

an hour ago

I predict once the responsible disclosure period is up we will see a lot more