Unverified50% confidenceFactExact time
Common sources of failure for desktop agents include: expired credentials or authentication failure, external service timeouts or rate limiting, empty or malformed input, and approval being rejected or no one responding.
1
Sources
50%
Confidence
Long-term
Relevance
10/15/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedNot knowing the Agent exists is the most common AI Agent security failure path, manifesting as failing to see that an Agent holds credentials it should not hold (such as service accounts, unapproved built-in tool permissions, or overly broad OAuth grants).64% similarUnverifiedThis update fixes the issue where sessions failed on every turn due to the API error "role 'system' must precede an 'assistant' message".63% similarUnverifiedPractices such as the 'must-fail test case' are also known in the industry as 'negative testing' or 'pessimistic path testing,' and are often mandatory requirements in certification standards for safety-critical systems (such as medical and aviation software), yet are frequently omitted in ordinary business development.63% similarUnverifiedFailure paths for gate rejection include: empty diff rejection, rejection for changes to test files only with no production code changes, rejection for production symbol changes without property tests, and rejection for properties passing on an unsealed summary.62% similarUnverifiedDevelopers should by default encapsulate business-logic-level errors (invalid parameters, non-existent resources, insufficient permissions, etc.) as tool errors; only genuine protocol violations should trigger a JSON-RPC error response.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1034343API
curl https://kongchang.com/api/v1/knowledge/claims/1034343MCP
get_claim(id=1034343)