On April 7th, Anthropic released a model that found thousands of zero-day vulnerabilities (some undetected…
Lara B. — Building Cyberr
On April 7th, Anthropic released a model that found thousands of zero-day vulnerabilities (some undetected for 27 years) and built working exploits on the first attempt 83% of the time.
They didn't release it publicly.
~40 organizations got access. AWS. Microsoft. Google. CrowdStrike.
The companies with the largest security budgets on the planet. The ones that needed it least.
Meanwhile, the organizations running the infrastructure that keeps society functioning - are still defending themselves with skeleton teams and quarterly patching cycles.
Anthropic's researchers estimate comparable capabilities will be available to other labs within 6 to 18 months. Potentially through open-source models.
So let's be clear about what that means:
Attackers will eventually have this. Most defenders won't.
That's not a cybersecurity problem. That's a civilization problem.
The question isn't whether AI changes the threat landscape.
It's whether we're building the future of defense for everyone - or just for the few who could already afford it.
4 reactions · 0 comments