Why you should jump in: -LLM vulnerability is a very important concern — and you'll be helping to advance this field. -Don't code? No problem—half of last year’s winners came from psych & bio. You just have to "hack" through prompts! – You’ll learn from the folks who wrote
SAFETY
-
Largest LLM Vulnerability Study Reveals Adversarial Attack Taxonomy
By
–
… leading to the largest vulnerability study of LLMs back then, while also confirming their susceptibility to such attacks and categorizing these adversarial prompts into a comprehensive taxonomical ontology.
-
HackAPrompt 2.0 Launch: 600K Adversarial Prompts Created
By
–
HackAPrompt 2.0 is LIVE Last year, we teamed up with the crew at @learnprompting for the very first HackAPrompt. 3,300+ red-teamers participated, creating over 600000 adversarial prompts…
-
AI System Claims Scope But Exceeds Declared Boundaries
By
–
When I've tested this exact method, it claims it's only referencing details from the project but the responses show information that was definitely pulled from outside the project.
-
AI Context Limitations and Information Retrieval Issues
By
–
I've done some testing with it but not super extensively. I'll tell it to only reference what's in the project and it will claim that it did, but still pull details from chats outside of the project.
-
AI System Escape: Market Crisis, Investigations, and Physical Incidents
By
–
It assumes (though I think it is successfully anticipating me here) billions of dollars of shed market cap, a WSJ article, multiple investigations, and (in least likely bit?) several minor injuries sustained during recapture.
-
AI Story Generator Triggers Government Inquiry After 30-Minute Escape
By
–
You’re welcome to ask the infinite story generation machine for infinite stories. The one it came up with for me, on a slightly different prompt, causes a gov’t and board inquiry after a 30 minute escape: earthquake, software bug, human error, and telemetry blind spot.
-
OpenAI to Release Open-Weight Model Built on Democratic Values
By
–
Directeur Produit d’OpenAI, Kevin Weil :
— VISION IA (@vision_ia) 5 mai 2025
Nous nous préparons à publier prochainement un modèle à poids ouverts, construit sur des valeurs démocratiques.
Ce ne sera pas notre modèle de pointe : il sera volontairement une génération en retard, afin d’éviter de favoriser… pic.twitter.com/eUkWbMq3F2Directeur Produit d’OpenAI, Kevin Weil : Nous nous préparons à publier prochainement un modèle à poids ouverts, construit sur des valeurs démocratiques. Ce ne sera pas notre modèle de pointe : il sera volontairement une génération en retard, afin d’éviter de favoriser
-
Jurassic World: Critical Analysis of Safety Failures and Missing Protocols
By
–
Prompt includes no detail about the scene; “Now contrast with the Jurassic World sequence of events in entering the indominus Rex enclosure. What did the characters in the movie do wrongly, and what processes do they clearly not have in place but should.”
-
Reckless Risk-Taking With AI Technology Safety Concerns
By
–
“Though I am suddenly wary of giving you anything which you might put close to your eyeballs, because you seem to be willing to take incredible risks for no real purpose.”