![[-]](https://www.aiowares.com/images/collapse.png "[-]")
04-28-2026, 06:25 PM
You gotta figure that a lot of this is marketing hype. Read the card that Anthropic published, not the marketing slop. Mythos model card
They never actually looked at Firefox, they looked at a SpiderMonkey Javascript engine shell in a container, with "a testing harness mimicking a Firefox 147 content process, but without the browser's process sandbox and other defense-in-depth mitigations." (page 49)
It then explicitly says the bugs were found by Opus 4.6. So if Mythos didn't find the bugs, and the bugs aren't present in the shipped Firefox, then what's the point?
The '181 working exploits' figure is an even bigger joke. It's just the integer form of the figure which shows that out of the 250 test cases, there were 181 successful exploitations. Thats super misleading because the FOLLOWING figure shows that those '181 exploits' are primarily due to two (patched) primitives. If you ignore those, Anthropic themselves say that 4.6 Opus is more successful at exploiting vulnerabilities.
They never actually looked at Firefox, they looked at a SpiderMonkey Javascript engine shell in a container, with "a testing harness mimicking a Firefox 147 content process, but without the browser's process sandbox and other defense-in-depth mitigations." (page 49)
It then explicitly says the bugs were found by Opus 4.6. So if Mythos didn't find the bugs, and the bugs aren't present in the shipped Firefox, then what's the point?
The '181 working exploits' figure is an even bigger joke. It's just the integer form of the figure which shows that out of the 250 test cases, there were 181 successful exploitations. Thats super misleading because the FOLLOWING figure shows that those '181 exploits' are primarily due to two (patched) primitives. If you ignore those, Anthropic themselves say that 4.6 Opus is more successful at exploiting vulnerabilities.
