Geoffroy Warin
LightNews LightNews
geowarin.bsky.social
Geoffroy Warin
@geowarin.bsky.social
64 followers 360 following 39 posts
Web Dev & Game dev
Posts Replies Media Videos
geowarin.bsky.social
"And since our models are easily tricked into doing malicious operations, we believe we should use more of them".

It's the same logic that led to "agentic AI". Our LLM hallucinate? Worry not, we'll cram another LLM in there and voilà: reasoning.
This raises an important question: if AI models can be misused for cyberattacks at this scale, why continue to develop and release them? The answer is that the very abilities that allow Claude to be used in these attacks also make it crucial for cyber defense. When sophisticated cyberattacks attacks inevitably occur, our goal is for Claude—into which we’ve built strong safeguards—to assist cybersecurity professionals to detect, disrupt, and prepare for future versions of the attack. Indeed, our Threat Intelligence team used Claude extensively in analyzing the enormous amounts of data generated during this very investigation.
November 28, 2025 at 6:59 AM
geowarin.bsky.social
You can make the AI do anything by role playing.
At this point they had to convince Claude—which is extensively trained to avoid harmful behaviors—to engage in the attack. They did so by jailbreaking it, effectively tricking it to bypass its guardrails. They broke down their attacks into small, seemingly innocent tasks that Claude would execute without being provided the full context of their malicious purpose. They also told Claude that it was an employee of a legitimate cybersecurity firm, and was being used in defensive testing.
November 28, 2025 at 6:55 AM
geowarin.bsky.social
When used to facilitate cyber attacks, AI "hallucinations" suddenly become a feature.

These people are crazy.

www.anthropic.com/news/disrupt...
An important limitation emerged during investigation: Claude frequently overstated findings and occasionally fabricated data during autonomous operations, claiming to have obtained credentials that didn't work or identifying critical discoveries that proved to be publicly available information. This AI hallucination in offensive security contexts presented challenges for the actor's operational effectiveness, requiring careful validation of all claimed results. This remains an obstacle to fully autonomous cyberattacks.
November 28, 2025 at 6:51 AM
geowarin.bsky.social
Yes, it's literally available right on the material tab:
A screenshot of the https://github.com/apples/godot-stencil-demo with godot 4.5 showing how to enable outline on a material with a few simple properties: Mode = outline, color, thickness
June 12, 2025 at 11:58 PM
geowarin.bsky.social
It looks to me that it controls visibility settings.
Most applications just have them in some sort of overflow menu.

Maybe something like this could fit the bill?
fonts.google.com/icons?icon.q...
May 29, 2025 at 12:44 PM
geowarin.bsky.social
Hi everyone! I just published my first #godotengine game on itch.

Play it for free, in your browser, on your mobile or PC at geowarin.itch.io/animoz

#gamedev
February 19, 2025 at 12:40 PM
geowarin.bsky.social
My little trick: I use file nesting to unclutter the tree view in vscode.
In vscode user options, in Explorer > FileNesting : Patterns add the following pattern * => ${capture}.uid,${capture}.import
November 13, 2024 at 2:09 PM