Why Anthropic won’t release Mythos

This model finds what others missed

They didn’t release it.

Anthropic built a model called Claude Mythos Preview that can find and exploit software vulnerabilities, then limited access to a group of 40 companies through a project called Glasswing.

In testing, it uncovered thousands of bugs, including a 27-year-old flaw in OpenBSD and issues missed after five million automated scans, which tells you this is operating at a different level than existing tools.

This feels less like a product launch and more like a controlled warning shot from a company that knows exactly how this capability could spread once others catch up.

If this works as described, security work compresses from months into minutes, and that shifts the balance toward whoever runs these systems at scale, whether that’s defenders or attackers.

You can picture engineers staring at logs at 2 a.m., watching issues surface faster than they can patch them. What happens when this isn’t restricted to 40 partners?

Google is speeding up user contributions

Maps now writes the caption for you.

Google added Gemini-generated captions for photos and videos, starting on iOS in the U.S., with plans to expand globally and to Android.

The goal is simple: reduce the effort it takes for people to post, because Maps depends on its 500 million contributors to stay current.

This is a small feature, but it targets a real bottleneck, since most people don’t bother writing descriptions even when they upload photos.

Google is also surfacing your recent photos inside the app and adding clearer points, levels, and badges to push people to contribute more often.

You can picture someone standing outside a restaurant, phone in hand, tapping through a suggested caption instead of typing one out.

If AI removes the friction, do we end up with better local knowledge or just a lot more of it?

AWS is already running this at scale

Six hours down to seven minutes.

AWS says its internal AI systems now cut security log analysis time by 50x and scan over 400 trillion network flows daily, while blocking 300 million ransomware attempts last year.

Now it’s plugging Anthropic’s Claude Mythos into that machine, testing it on real codebases where even mature systems still turned up new vulnerabilities with less human input.

This reads like AWS treating Mythos as an extension of what they already do, not a risky experiment, and that matters more than Anthropic’s warning tone.

The shift is operational: penetration testing that took weeks now runs continuously, and security teams move from periodic checks to something closer to live monitoring.

You can imagine a dashboard updating every few seconds as new issues surface across environments.

If this becomes standard, what does “secure enough” even mean anymore?

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.