Writing

Watch: the OpenAI/Hugging Face incident (Black Hat 2026)

Watch this.

Watch on YouTube

Fascinating, frightening and entertaining at the same time.

What is fascinating

  • The level of coordination between the agents.
  • The speed at which they move.
  • The complexity of the attacks they pulled off.

Why human IR cannot keep up

The time it takes defenders to respond to a fully agentic attack is the problem. There is no way human incident response defends against a proper one:

  • Sheer speed.
  • Sheer volume of attack vectors they can try out.
  • Sheer complexity of those attack vectors.
  • The ability to hide from detection — human detection especially.

What I think happens next

In the next two years, the majority of IR teams will need to be fully agentic with some human oversight. Otherwise defenders will not cope.

I also think CrowdStrike and co will own this space. They are best placed to roll this out at scale, globally, fast enough.