Claude Mythos: Anthropic Bans AI That Hacked OpenBSD and FFmpeg
Tech News

Claude Mythos: Anthropic Bans AI That Hacked OpenBSD and FFmpeg

6 min read
77 Views
Share:

Anthropic stops the launch of its most powerful model

Anthropic made a historic call this week: banning public access to Claude Mythos Preview, a model with coding and cybersecurity abilities so advanced the company itself deems it a global security risk. Having covered AI launches since 2022, this is the first time a company has voluntarily halted a model out of fear of its own creation.

The news broke on April 11 and has already sparked emergency meetings across Anthropic, the White House and Wall Street.

What Claude Mythos did to scare everyone

The model discovered two critical vulnerabilities that had remained hidden for decades:

  • A flaw in OpenBSD undetected for 27 years.
  • A bug in the FFmpeg video library that survived 16 years and over 5 million automated tests.

On benchmarks, Mythos posts numbers no previous model has reached:

BenchmarkClaude Opus 4.6Claude MythosGPT-5.3
SWE-bench Verified74.5%93.9%82.1%
USAMO (math olympiad)83%97.6%89%
0-day discoveryLimitedAutonomousLimited

Government and market response

Treasury Secretary Scott Bessent and Fed Chair Jerome Powell called an urgent meeting with financial-system leaders. The reason: if such a model leaks, it could hit banking infrastructure within hours.

Wall Street reacted instantly. Traditional cybersecurity stocks fell hard:

  • CrowdStrike: -11% at the open
  • Palo Alto Networks: -8%
  • Fortinet: -5%

How it affects you

Even though you will not get to test Mythos, its consequences reach you:

  • Urgent patches: expect critical updates for OpenBSD, FFmpeg and systems built on them (most Linux servers and video apps).
  • Cybersecurity pricing: vendors will accelerate investments in AI-assisted detection.
  • Regulation: a new legal framework around offensive-capable models is on the horizon.

Command to check your FFmpeg version

ffmpeg -version | head -n 1
# If you see a version older than 7.1.2, upgrade:
sudo apt update && sudo apt install --only-upgrade ffmpeg

Troubleshooting

"My server runs OpenBSD — am I at risk?" If you run a version older than 7.6, apply the upcoming security patch. In the meantime, monitor logs with tail -f /var/log/authlog.

"I use FFmpeg for streaming, should I shut it down?" No, but upgrade to the latest stable build and check the official security advisory every 48 hours.

"How do I get notified of the official patch?" Subscribe to the OpenBSD and FFmpeg security lists. I have relied on them for years and they have never failed to warn in time.

What comes next

Anthropic already opened the "Glasswing" program so security researchers can access Mythos under strict agreements. After seeing this dynamic with prior models, my read is that some capabilities will eventually reach Claude Opus, but wrapped in usage safeguards.

Additional resources

J
Written by
Jesús García

Apasionado por la tecnologia y las finanzas personales. Escribo sobre innovacion, inteligencia artificial, inversiones y estrategias para mejorar tu economia. Mi objetivo es hacer que temas complejos sean accesibles para todos.

Share post:

Related posts

Comments

Leave a comment

Recommended Tools

The ones we use in our projects

Affiliate links. No extra cost to you.

Need technology services?

We offer comprehensive web development, mobile apps, consulting, and more.

Web Development Mobile Apps Consulting