@andrei_chiffa@mastodon.social
Post #2459773
2026-04-27 16:08 UTC
For those who have followed me here or have heard my talks, the take-out will be unsurprising:
- Mythos is but a sign of a larger trend of LLMs trained into cyber-offesnive
- Such capabilities in public models is asking for trouble; Advanced Persistent Teenagers getting exploits left and right is a good way to get Not Petya repeat as many times as it takes to break the world's IT infrastructure
- Security by obscurity - advocated fro by Anthropic - is a bad idea.
2/🧵
Replies (1)
-
@andrei_chiffa@mastodon.social 2026-04-27 16:09
- We need pre-release standards and third-party testing - Finally, we need proper dissuasive legislation to enforce adherence to the most basic cybersecurity and AI safety standards from LLM developers, starting with putting the bill for incidents to the LLM developers. 3/🧵