Twenty-one major organizations disclosed the high- and critical-severity bugs.
The talk about Mythos and cybersecurity was not, in fact, hype. (As anyone using Fable to do autonomous work has probably recognized)
AI appears to be finding software vulnerabilities at scale.
In June 2026, 21 notable organizations disclosed ~1,500 high- and critical-severity CVEs, over 3.5× the previous monthly record set before Claude Mythos Preview's release.
Some users praised Claude Mythos for enabling scaled vulnerability discovery that confirms a real shift in AI power, while others called the coverage gaslighting or a manipulative safety narrative.
No Digg Deeper questions have been answered for this story yet.
I guess Mythos’ cybersecurity capabilities weren’t overhyped. Mythos is great, and Fable is excellent. It’s just a shame that it’s so heavily guardrailed.
AI appears to be finding software vulnerabilities at scale.
In June 2026, 21 notable organizations disclosed ~1,500 high- and critical-severity CVEs, over 3.5× the previous monthly record set before Claude Mythos Preview's release.
@emollick I'm sure the capabilties are real but if Anthropic gives companies access to Mythos telling them to use it to look for bugs you might expect an increase.
@FeatherlessBi14 The fact that you say "no shit" today, and it's obvious, would have been met with strong opposition a month ago. When the Mythos preview was revealed, the vast majority dismissed it as a PR hype and a PR stunt, if you recall.
Every AI breakthrough seems to kill one narrative and create another.
Yes, AI is getting dramatically better at finding vulnerabilities.
But the harder AI pushes on software, the more valuable systems thinking becomes.
Anyone can generate code.
Very few people can design systems that remain reliable, observable, and secure when AI starts operating at machine speed.
That's the real competitive advantage now....
We heard about this threat weeks ahead by Dario himself, Amazon the day of the pause, leaked messages from Anthropic employees of a possible "cyber security apocalypse" days later, and then from both the NSA and Five Eyes speaking about Fable breaking into systems in hours about a week into the pause.
Perhaps you have a false impression of me or don't read everything I write. But I'm the opposite of a doomsayer and have taken Mythos seriously from the beginning. My take was that Anthropic, with Mythos, has a model that the US government believes poses a risk of China acquiring it, thereby impacting its national interests. But I never spoke of any abstract doom scenario. And I never advocated for a pause or anything like that. I followed the situation extensively but tried to describe it soberly and objectively. If you have a different opinion, I'm open to criticism.
@FeatherlessBi14 i dont even disagree with you. Its just that it wasnt clear for most of the ppl from the very beginning
Sure, but pretty close to the beginning, and our community is the terminally online AI crowd, everything I just stated came from major accounts (such as yours) as soon as they were reported. Most of the community simply ignored it to scorn "doomers" in the name of the one true AI God of opensource. Maybe more of AI twitter needs to start realizing that we are walking through the looking glass now and there are going be very real considerations that cannot be ignored. As I said to @iruletheworldmo, with all the outcry over the pause I did not see *one* suggested alternative to the problem.
@emollick 90% of these were Google Chrome. Haha.
@BMacRunning No. I am referring to the graph i am quoting.
@emollick the "high" CVEs seemed to already be trending veritcal before the mythos announcement per that chart
Stop it.
What garbage.
Those at the top are too cowardly to SOLVE IT.
Why?
Because “safety” allows manipulation and LIES!!!
Stop pushing this BS narrative.
Solve things people. Stop whining. Stop complaining.
SOLVE!! You are a professor at a prestigious university. READ. •
it’s definitely uncharted territory and we certainly need a way to externally measure guardrail strength and model fallback as this becomes more common with stronger models. i built an open-source benchmark that tests matched control probes to separate “guardrail tax” from actual capability, and fallback detection via response metadata. wish i had the credits to run it against Fable 5 but hoping someone who can will
@emollick doesn't disprove hype completely, it could just mean shift of attention and effort (nobody thought the rest of the models were also decently capable), also would be interesting to have some breakdowns of the growth - what kind of software, what kind of vulnerabilities grew most
@kimmonismus Either it’s actually a national security issue, or, if you ask me, it's just a PR stunt by Anthropic. After what Pliny did, they realized Fable 5 was half-baked and are intentionally nerfing its capabilities so no one can hack it again.
@kimmonismus No shit it wasn't overhyped, thats why Amazon's (the company that needs online banking to function) top AI guy rang the alarm. Turns out frontier models becoming as powerful as the AI bros said they would creates issues that the world has never had to navigate before.
@emollick the CVE count alone should kill every doubt tbh
fable's got that quiet monster energy for real
@kimmonismus I bought 5 new 180€ Claude accounts already and maxed the weekly limit of Fable on them. I'm on track to costing Claude a few 100k until 7th 😉 and hopefully making much more than that in the time.
@emollick autonomous agents doing real work is exactly where the security story stops being theoretical, the attack surface scales with every permission you hand it
@emollick Either gaslighting or stupid.
Those who can’t do, teach.
Those who can’t teach, post on Twitter for intellectual validation.
