AI cyber risk is entering a new phase, and Anthropic’s Mythos may prove to be a watershed moment for cyber security.
Anthropic’s newly announced Mythos Preview model has not been released publicly. Instead, it has been restricted to a select group of major technology and financial organisations under its Project Glasswing initiative.
Why?
Because according to Anthropic’s own testing, Mythos demonstrated the ability to autonomously identify and exploit serious software vulnerabilities at a level far beyond previous frontier AI models.
If accurate, this represents more than another AI announcement. It signals the beginning of a new era in cyber security, one where AI powered threats may accelerate faster than many organisations are prepared for.
What Is Anthropic’s Mythos?
Mythos is Anthropic’s latest advanced language model, reportedly developed as a general-purpose AI system.
However, during internal testing the model displayed an unexpected capability:
It became exceptionally effective at identifying vulnerabilities and developing working exploits against real software.
Anthropic claims Mythos can:
- Discover zero-day vulnerabilities in major software platforms
- Turn known vulnerabilities into working exploits far faster than previous models
- Chain multiple vulnerabilities together into sophisticated attack paths
- Assist even non-specialist users in vulnerability discovery and exploit development
This capability was significant enough for Anthropic to withhold public release.
Why Project Glasswing Matters
To manage the risk, Anthropic launched Project Glasswing, a controlled initiative allowing selected organisations to use Mythos defensively.
Participants reportedly include major banks and cyber security vendors.
The purpose is simple:
Use advanced AI to help defenders find and fix vulnerabilities before similar capabilities become available to attackers.
That alone should tell organisations everything they need to know.
If frontier AI labs are limiting access to a model because of cyber offensive capability, the threat landscape is changing.
Why This Changes the AI Cyber Risk Conversation
Until now, much discussion around AI powered cyber-attacks has focused on:
- Better phishing emails
- Faster malware variants
- Improved reconnaissance
- More convincing social engineering
Mythos suggests the next evolution may be far more serious.
AI Is Moving Up the Attack Chain
Rather than simply assisting with peripheral attack tasks, advanced models may now help with:
- Vulnerability discovery
- Exploit development
- Attack path chaining
- Offensive security automation
In practical terms, that means:
The barrier to sophisticated cyber-attacks may fall dramatically.
Why Businesses Should Pay Attention Now
Many organisations still assume sophisticated exploit development is reserved for elite attackers.
That assumption may no longer hold.
If AI can meaningfully accelerate vulnerability exploitation:
Lower Skill Attackers Become More Dangerous
Less experienced threat actors may be able to:
- Exploit vulnerabilities faster
- Attack more complex systems
- Scale campaigns more efficiently
- Weaponise disclosed CVEs rapidly
Patch Windows Become More Critical
Anthropic specifically warns that:
Time between vulnerability disclosure and exploit weaponisation may shrink significantly.
This makes slow patching increasingly dangerous.
What Mythos Means for Cyber Defenders
The key takeaway is not panic.
It is preparation.
The organisations most exposed in an AI accelerated threat landscape will be those with weak fundamentals.
That includes businesses with:
- Poor vulnerability management
- Slow patch cycles
- Legacy unsupported systems
- Limited visibility across their estate
- Under resourced detection and response capability
How Organisations Should Respond to Rising AI Powered Threats
Strengthen Vulnerability Management
Know what assets you have, where vulnerabilities exist, and how quickly they are being remediated.
Reduce Patch Latency
Faster attacks require faster detection and response.
Focus On Cyber Resilience
Prevention remains essential, but resilience matters more than ever.
Businesses must prepare for:
- Rapid containment
- Effective recovery
- Tested incident response
- Business continuity under attack
For More Information
Mythos Is Likely Only the Beginning
Anthropic itself states that Mythos is unlikely to represent the ceiling of AI cyber capability.
In fact, they believe future models will continue improving rapidly.
That means the real question is not:
“Will AI change cyber security?”
It already is.
The question is: “How quickly will organisations adapt?”
Final Thoughts on AI Cyber Risk
Anthropic’s Mythos and Project Glasswing may be remembered as the moment AI powered threats moved from theoretical concern to practical reality.
Whether Mythos itself becomes widely available is almost secondary.
The capability exists.
Others will follow.
For defenders, the lesson is clear:
The age of AI accelerated cyber risk has begun. Organisations that improve visibility, reduce exposure, and build resilience now will be best positioned for what comes next.
Call UsFor More Information
Need Help Preparing for Emerging AI Cyber Risk?
At Amicis Group, we help organisations strengthen cyber resilience through proactive security monitoring, vulnerability management, incident response and managed cyber security services.
If your organisation wants to assess its readiness for the next generation of AI powered threats, speak to our team.
Call us today on 0333 305 5348 or email hello@amicisgroup.co.uk to talk to us about your cyber security.
You may be interested in our article on CrowdStrike’s Global Threat Report 2026 or our SOC cyber security information article.
