it’s the 1990s again in infosec/security

Our evaluation of Claude Mythos Preview’s cyber capabilities | AISI Work We conducted cyber evaluations of Anthropic’s Claude Mythos Preview and found continued improvement in capture-the-flag (CTF) challenges and significant improvement on multi-step cyber-attack simulati...

My guy I literally sent you a benchmark from the UK's AI security institute to prove my point. I do not give a fuck if you believe me or not😂. I block idiots.
www.aisi.gov.uk/blog/our-eva...

t.co/TMCEjEr1KK

Well... it's not just my opinion and I'm not just some rando Cyberpunk pfp lmfao. I'm actually one of the highest ranked ethical hackers on a platform as we speak.

You don't need to take my word for it though. From the AISI's benchmark.

And I don't just mean on detecting vulnerabilities in code. I mean in live environments.

It's very wide actually. It might be closing(If and when OAI chooses to provide access to their cyber fine-tuned models to individuals) but as is, Anthropic has the best models for infosec use cases by a mile.

All the best hackers in the world already know. It's not a vibe...

Contra Benn Jordan, data center (and all) sub-audible infrasound issues are fake One of the most popular videos made about data centers ever is a complete moment-by-moment disaster

Folks, infrasound issues are fake. This was truly an insane experience to write and I hope you enjoy blog.andymasley.com/p/contra-ben...

Scoop: NSA using Anthropic's Mythos despite Defense Department blacklist The government's cybersecurity needs are outweighing the Pentagon's feud with Anthropic.

sucks to be this good

www.axios.com/2026/04/19/n...

Palantir: we will make you all serfs in our new dawn techno-fash state

Pro tech people: this is actually just the normal American RW talking, freedom of speech yk

Normal people: ok, I think Palantir investors should be shot

Pro tech people: nooooooo this is violence, commies can't stand freedom

Lmfao.

know*

A harness that says, "inspect this code for memory corruption bugs" will throw a shit ton of false positives your way, which is what the Open Source studies that came out missed.

The reason I know how good Mythos is, is because it's an inductive proof of on what i know to be hands-on capacity. My VR and Binary exploitation friends will tell you as much too.

The Open Source study you cited was bad for several reasons no less because we rarely no what we are looking for.

I pentest as a hobby and I've used Opus 4.5 to take over the broadcast infra of a major media org. I have empirical experience on how good the various SOTA are. You are not doing what I did with any model that isn't a Claude. Increasingly a Gemini somehow.

The Kimis,GLMs,etc are getting good tho.

Dr Kareem Carr @kareem_carr · 4h There's a bias in AI discourse where we say certain things were achieved with AI, but they were also achieved with millions of dollars. Like if we gave every traditional scientist tens of millions per year, they'd probably be coming up with plenty of fancy new solutions too.

I used to believe this and then I saw the budget requested for a traditional chemistry grant.

Interesting. Claude Code's default model has been changed to Sonnet 4.6 from Opus 4.6 with 1M context, which became extra usage.

Gary Marcus building his case that the Top Secret list of NSA 0-days leaked into the training data.

Mythos

this is about my posts btw

My quantisation post is doing the rounds again over in the Elonosphere and it feels like about 80% of the replies/comments are AI generated.

Some of the replies I suspect are human give themselves away by asking questions answered near the end of the post. At least the LLMs read the whole thing.

It doesn't help that this creator specifically consistently downplays great breakthroughs.

You have to understand that there's a section of people who pride themselves on the belief that none of the developments taking place in AI are worth paying attention to and so the tone here is precisely crafted to soothe that audience.

The advisor strategy on the Claude Platform

We're bringing the advisor strategy to the Claude Platform.

Pair Opus as an advisor with Sonnet or Haiku as an executor, and get near Opus-level intelligence in your agents at a fraction of the cost.

Melania Trump Denies Ties to Epstein in Rare Public Statement First lady Melania Trump denied ties to the late, disgraced financier Jeffrey Epstein, calling online claims about a supposed relationship “false smears” and threatening retaliation against those making them.

First Lady Melania Trump denied ties to disgraced financier Jeffrey Epstein and called for a congressional hearing to allow survivors of his abuse to tell their stories.

EFF is Leaving X After almost twenty years on the platform, EFF is logging off of X. This isn’t a decision we made lightly, but it might be overdue.

After almost twenty years on the platform, EFF is logging off of X.

This isn’t a decision we made lightly, but it might be overdue. 🧵 (1/5)
www.eff.org/deeplinks/2...

Nope. Even basic Claude models were highly capable wrt this stuff. Mythos is obviously a leap step above current SOTA. But Anthropic already had the best models for infosec.

through years of LLM research, we have finally managed to invent a human-usable interface for ffmpeg

People say ‘oh we just replaced one Ayatollah with his son that’s not really regime change’ but actually it looks like we probably replaced a semi-constitutional theocracy with a military junta + religious figurehead