@cR0w ugh. this is how “they”’re trying to defend “agents” https://www.philschmid.de/why-engineers-struggle-building-agents
undefined
hrbrmstr 🇺🇦 🇬🇱 🇨🇦
@hrbrmstr@mastodon.social
Posts
-
@cR0w ugh. -
"AI" is going GREAT!"AI" is going GREAT!
"Second-order prompt injection attacks" in agent-to-agent setups. Demonstrated with ServiceNow.
https://appomni.com/ao-labs/ai-agent-to-agent-discovery-prompt-injection/
A2A was released in April 2025 and we already have not only direct and first-order attacks, but also, now, second-order ones.
Non-deterministic processing meets impossible-to-ensure-safe-and-resilient workflows.
WCPGW?
-
Microsoft steals content for their AI training data but doesn't want others to do the same thing.Microsoft steals content for their AI training data but doesn't want others to do the same thing. Odd.