AE-DVYMQV5D
Model hallucinates a call to a native tool that was not enabled; replaying it on retry makes the provider reject the request with HTTP 400
HALLUCINATIONseverity: MEDIUMcause: LIKELYoutcome: RESOLVED UNVERIFIEDconfidence: LOW
With pydantic-ai-slim[anthropic] 2.7.0, Claude occasionally returned a server_tool_use block for a native tool (code_execution) that was not enabled for the request. On the validation retry the agent replayed that call verbatim without a result, and the Anthropic API rejected the next request with HTTP 400, failing the run.
- Framework / agent
- Pydantic AI (pydantic-ai-slim[anthropic]) · Pydantic AI agent on Anthropic
- Remediation attempts
- TESTEDSUGGESTED
- Recurrence
- not documented
- Source languages
- en
- Updated
- 2026-09-29
Sources
- GITHUB ISSUE Gracefully handle hallucinated Anthropic native tool calls — github.com/pydantic/pydantic-ai, retrieved 2026-09-29
- GITHUB PULL REQUEST Filter unconfigured Anthropic native tool calls — github.com/pydantic/pydantic-ai, retrieved 2026-09-29
- GITHUB PULL REQUEST fix(anthropic): drop hallucinated native tool calls for disabled tools (#6401) — github.com/pydantic/pydantic-ai, retrieved 2026-09-29
Symptoms
- The model returns a native code_execution tool call although no tools were configured
- The retry request is rejected with HTTP 400 because the replayed tool use has no matching result block
The full record — root-cause evidence, every remediation attempt with its status and verification, failed attempts, patch references, verbatim quotes and recurrence — is a paid lookup (0.018 USDC via x402). Agents:
GET /api/v1/cases/AE-DVYMQV5D. PricingSimilarity to your system is not implied. A remediation that worked in the documented context may not work in yours.