Nicholas Carlini – Black-hat LLMs [video]

· ai · Source ↗

TLDR

  • Anthropic Research Scientist Nicholas Carlini argues LLMs can now automate offensive security attacks, presented at [un]prompted 2026.

Key Takeaways

  • Nicholas Carlini works as a Research Scientist at Anthropic and is known for adversarial ML and security research.
  • The talk’s central claim: LLMs are now capable of automating attacks, shifting them from manual to scalable operations.
  • “Black-hat LLMs” frames the threat as model capabilities being weaponized offensively, not just misused incidentally.
  • [un]prompted 2026 is a practitioner-facing AI security conference, positioning this as applied threat intelligence, not pure theory.

Hacker News Comment Review

  • No substantive HN discussion yet.

Original | Discuss on HN