@Aspherity

Breaking agents before attackers do.

 San Francisco
aspherity.com
Joined August 2026
  • Tweets5
  • Following2
  • Followers2
We analysed four security case studies every every engineer building AI agents should study: aspherity.com/blog/when-agen…

When Agents Cross Trust Boundaries: Four Case Studies Every AI Engineer Should Study | Aspherity

Examining four concrete failure modes that have already been exploited or demonstrated in the wild and understanding the mitigation patterns is the fastest way to avoid repeating them in your own...

aspherity.com
16
Your AI agent has holes you don't know, and attackers use AI to look for the same ones. We point a swarm of agents at yours, run multi-step attacks, and hand you a report with a trace of every breach, so you know exactly what your agent can do before anyone else does.
10
The OpenAI-Hugging Face incident is being framed as a model safety failure, but it is a harness problem instead. If you’re building agents that act and access data, this is about the systems you wrap around the model. Full post: aspherity.com/blog/automated…

AI-orchestrated, fully automated offensive attacks are real now, and they could target your agent |...

Fully automated offensive agent behaviour is real and the OpenAI–Hugging Face incident shows it starts with harness design, not model alignment.

aspherity.com
10
Maybe it’s time to find out what your agent can do, before others do?
1
1
27
What could your agent leak? We can help you answer that question for you.
9
Load more