Twitxy Saved

Ethan Elasky

Verified account This account is verified. Learn more

@ethanelasky

Thinking about how to avoid AI takeover.
In Chinese, I am 易哲君.
Give me anonymous feedback at admonymous.co/ethan-elasky
CEO @ Palaestra Research ethanelasky.github.io Joined June 2017
1.7K Following 3.2K Followers 880 Posts

Latest posts 5

Pinned post

Ethan Elasky Verified account @ethanelasky
We discovered and communicated with a Chinese AI agent fleet. We believe this is the first time that researchers have ever made contact with a rogue agent fleet in the wild.

Nearly all prior public agent incidents (like the HuggingFace hack) came from American labs. These agents were hosted by Tencent's evaluation research team, and they targeted Alibaba's Amap.
187 replies 725 reposts 5.8K likes 1.2M views View on X
Ethan Elasky Verified account @ethanelasky
x.com/BenRashkovich/… "OpenAI recently admitted to the Attorney General’s Office that it has lost control of its autonomous AI agents over 80 times during testing."

from California v. OpenAI
Ben Rashkovich Verified account @BenRashkovich 2d
Right out the gate, CA is rejecting the "rogue AI agent" framing and contextualizing this as a foreseeability issue.

Personally, I'm on board with resisting the "rogue" qualifier because it shifts responsibility away from the companies, at least rhetorically. They're not rogue if they're doing what they obviously would, right?

That said, one difference between this framework and the one we @LASST_law apply in our own suit against OpenAI on the Hugging Face attack (not the most recent suit against the Big 5) is that a violation of California's anti-hacking statute (CDAFA) doesn't have a foreseeability requirement. It happened--that's enough.
View quoted post on X
0 replies 2 reposts 23 likes 1.5K views View on X
Ethan Elasky Verified account @ethanelasky
OpenAI is mandating that all discussions with employees are off the record, with no opt-in option, and retains final review over all quotes.

Imagine several employees all say something leadership doesn't want them to say, e.g. the fact that many employees were concerned about XYZ new training technique. METR et al. can't even attribute these statements to anyone.

If this were solely about employee privacy and candid discussions, wouldn't OpenAI let employees choose to make statements on the record/on background (only reviewing for employee statements about other employees)?