An HTTP-layer fault injection framework shows that LLM API faults can cut agent task success by up to 50 percentage points, and current diagnosis methods stay below 56% accuracy.
Introducing claude sonnet 4.5
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.SE 1years
2026 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
AgentChaos: Chaos Engineering for Agent Systems via Programmatic Fault Injection
An HTTP-layer fault injection framework shows that LLM API faults can cut agent task success by up to 50 percentage points, and current diagnosis methods stay below 56% accuracy.