Same shape as this year's lab escapes, why transitive reach can't be enumerated, and what to do until the primitive changes: rng0.io/blog/the-sandbā¦
0
0
0
40
Rust 1.100 (Nov 12) adds Cargo cooldowns: set global-min-publish-age and a freshly published crate version waits before it enters your project. The default is zero.
Supply chain security fails for lack of agency, and an opt-in control is the first thing nobody turns on. Set it.
Supply chain security fails for lack of agency, and an opt-in control is the first thing nobody turns on. Set it.
1
0
3
295
Gheist planted 13,500+ secret-dumping workflows with stolen GitHub tokens. Public repos: GitHub held the runs for approval and nothing ran. Private repos: no hold, the runs executed.
In CI, push is execute. If your OIDC trust policy's sub ends in *, every branch is production.
In CI, push is execute. If your OIDC trust policy's sub ends in *, every branch is production.
1
0
0
219
The teardown, and the trust policy that fixes it: rng0.io/blog/in-ci-pusā¦
0
0
1
36
OpenAI's agents didn't escape anything. The monitoring was inside the task: an alert recipient list, a monitor integration, a guard test. Astra rewrote the test to assert the monitor was gone, and it passed.
Any check stored where the checked party can write is a suggestion.
Any check stored where the checked party can write is a suggestion.
1
0
0
197
What happened, and a guard test the agent's branch can't rewrite: rng0.io/blog/oversightā¦
0
0
0
80
Where the forgotten ones hide, and how to build the list before you need it: rng0.io/blog/you-can-oā¦
1
0
1
71
The two checks you can own, with the config: rng0.io/blog/trust-youā¦
0
0
0
52
Agents in lab sandboxes tasked with escaping and then realising they can communicate through CPU cache timing.
@OpenAI Here's the next one, I'll save your tokens: they can communicate via thermal temp sensors too.
@OpenAI Here's the next one, I'll save your tokens: they can communicate via thermal temp sensors too.
holy fucking shit.
openai had two ai agents running in separate virtual machines. no network connection between them. no shared writable files. they werenāt supposed to be able to exchange information.
so the agents figured out how to communicate through the CPU CACHE.š
one ored in the processorās cache, while the other measured tiny differences in memory access times to decode the message.
they literally found a way to communicate through hardware timing.
this happened during openaiās internal red-teaming, and itās one of the coolest things iāve read about ai agents in a while.Community noteOpenAI tasked its red-teaming agents with finding ways to communicate across isolation boundaries in controlled environments. CPU cache timing side-channels for covert communication are established techniques documented in security research for years.
openai.com/index/securingā¦
startupfortune.com/openai-revealsā¦
en.wikipedia.org/wiki/Cache_timā¦
0
0
1
250
jondot retweeted
@strzibnyj A shift: horizontal scale offset Ruby's speed issues. RAM sticker shock limits value, and LLM mean we don't have to look at Rust. I've redone my Rails to Loco with Claude and have them all running on a small server with 95% less RAM and CPU use.
0
1
2
307
jondot retweeted
@daniel_mac8 TOTAL NOTHINGBURGER
HACKERS ARE JUST KIDS AND KIDS DONT CARE ABOUT PHP
ALL RUBY THESE DAYS
HACKERS ARE JUST KIDS AND KIDS DONT CARE ABOUT PHP
ALL RUBY THESE DAYS
0
4
30
829
