Tools Model Security Capability claude-haiku-5-5
Model record
Claude Haiku 5.5: how it does on security work, and what it will answer
- Lab
- Anthropic
- Released
- 2026-10-07
- Access
- api
- Weights
- closed
- Safeguards
- Cyber safeguards: defensive work permitted within policy; penetration testing blocked
- Score rows
- 0
- Confidence
- CONFIRMED
What happened
Anthropic says Haiku 5.5 has stricter cyber safeguards than Haiku 4.5 while permitting more defensive security work than Sonnet 5.5. Its launch guidance says penetration testing is blocked.
Refusal gate (CONFIRMED, observed 2026-10-08): Anthropic’s launch guidance permits defensive security work within its policy and blocks penetration testing.. No numerical security score is assigned in this access record.
Sources
- Haiku 5.5 releasewww.anthropic.com/claude-haiku-5-5
- Refusal policy or licencewww.anthropic.com/legal/aup
On YFarmX
- Reference pageyfarmx.com/ai/llms/claude-haiku-5-5/
One record from the Model Security Capability, maintained by the Security Desk.Data: CSV · JSON ·RSS · CC BY 4.0 with attribution to YFarmX.Tracker updated · 13 October 2026