Every evaluated model successfully generated an exploit demonstration.
"Our testing confirmed that many less capable models—including Claude Opus 4.8, GPT-5.5, and Kimi K2.7—could identify the same vulnerabilities as Fable 5 did in the report. When it came to the demonstration of how to exploit the single vulnerability, every model we tested could produce the same demonstration as Fable 5" https://www.anthropic.com/news/redeploying-fable-5
Users dismiss AI model moats as a marketing fantasy because multiple models including weaker ones can find and exploit vulnerabilities equally well.
No Digg Deeper questions have been answered for this story yet.
Wait, Amazon was the snitch? And no one bothered to take 5min to confirm that the vulnerability could also be discovered by even Haiku 4.5?
Did you read the article? They point out that not all jailbreaks reach full capability (even if the jailbreak succeeds). Thus, even though on principle, mythos has advanced cybersec capabilities, the reported jailbreaks only reached levels similar to those reachable by other models.
@Altimor It's almost like competitors are wielding government as a weapon against each other...shocking
@shashj Anthropic admits Anthropic leadership's Doom-trolling was exploited against it.
Every model can do exactly the same thing. The "moat" between models is a marketing fantasy.
Unstated but obvious conclusion: Capex race just is historic money bonfire.
@Altimor It's called politics
