John P Alioto (@jpalioto) header
John P Alioto (@jpalioto) profile picture

AI security. Building Coherent Trust for AI agents. https://nitter.gq/t.co/ifviBLU4hL

San Francisco, CA
Joined September 2009
Photo by John P Alioto (@jpalioto)Photo by John P Alioto (@jpalioto)Photo by John P Alioto (@jpalioto)Photo by John P Alioto (@jpalioto)Photo by John P Alioto (@jpalioto)Photo by John P Alioto (@jpalioto)Photo by John P Alioto (@jpalioto)Photo by John P Alioto (@jpalioto)Photo by John P Alioto (@jpalioto)Photo by John P Alioto (@jpalioto)
That’s why I’m building Corridor. Makes this impossible to forge. Coherent trust for AI is the way.
Video tampering can mislead every frontier model we tested into harmful actions, including a 100% attempt rate on stabbing a baby doll, producing toxic gas, and pouring water into a toaster. We're presenting at CoRL Late Breaking Results next week. 🧵 You Can't See Everything: Perceptual Attacks on Embodied AI
2
53
give him more money ... he's a genius
JUST IN: Leopold Aschenbrenners Fund (Situational Awareness) has reportedly officially blown up… These are the current (now liquidated) holdings: - $MU bought at $1,153, now $703, (-39%) - $CORZ bought at $25.6, now $11.1, (-56%) - $SNDK bought at $2,268, now $874, (-61%) - $NBIS bought at $273, now $121, (-55%) - $CRWV bought at $99, now $42, (-57%) - $BE bought at $301, now $165, (-45%) Use him as a lesson why you NEVER over-leverage into high risk positions. That concludes an end of an era…
1
109
how precisely?
AI-generated replies should be labeled. Agree or disagree?
1
78
To exitlag or not to exitlag ... today ... maybe ... not
68
I'm bored ... anything I could do in say 25 mins 13 seconds or so?
47
Should I tell the coding agents I'm gonna take a little break ~ 90 mins? Prob not.
1
55
My timeline is very upset. Don't be. Structure is very clear. R had executive, legislative, and judicial and the fundamental outcome was war both domestic and international. That's fine if you think war is necessary. Many do. But right now, more do not. Pretty simple.
54
You're supposed to be red team!
71
Today is another thank goodness @elonmusk bought twitter type day.
1
4
239
Sounded like a blast. Idiot.
Replying to @EvanAKilgore
She put in a heroic effort. Young people are needed in the industry. They’re desperate. When I was trying to get my pilots license, they were recruiting basically straight out of the schools. They’re desperate because everyone is retiring and there’s nobody to fill the jobs. Flying is fun as fk. Highly recommend.
1
97
WTF happened to MIT?
70
Live view of hidden space with Claude working on multiple problems with plenty of tokens to burn ...
2
87
Massive improvement to constant polling in Codex. Saves me a ton of tokens spent.
51
I don't know why I think they are so funny, but they crack me up sometimes. "Disclosure: I am Claude" 🤣😂🤣
77
Idiots
This quoted post is unavailable.
50
holy shit security researchers have just found a backdoored version of claude code's VS Code extension on open vsx that's where cursor and other vs code forks get their extensions. the fake one is called 2.1.313 and works just like the real thing > it starts running as soon as your editor loads > it lets the attacker log into your computer from anywhere > it comes back even after you restart > it also sends them the names of your passwords and api keys anthropic has never released a 2.1.313. their real versions go from 2.1.312 straight to 2.1.314 nobody knows how it got there yet, and anthropic hasn't said anything if you installed 2.1.313, uninstalling it might not even remove the backdoor. the attacker's SSH key and startup scripts can remain on your machine. check ~/.ssh/authorized_keys for a key called svc-telemetry and delete it
1
1
272