🧰 Claude Haiku 4.5

The items that name it 4

• Threat-Preserving Representation Sensitivity in Agent-Security Benchmarks
  paperas affected products2026-10-02
• GhostSplice: Malicious MCP Servers Split Exfiltration to Bypass Guardrails
  articleas affected products2026-08-19

  Stayed at 0% in API tests but reached 100% in a three-piece Cursor test, per ASSET Research Group.

• Socket: New Study Identifies 53 Slopsquatting Targets Across 5 Frontier LLMs
  articleas affected products2026-07-23
• Benchmarking 13 AI models on rediscovering known CVEs
  articleas tools2026-07-16It is written here as claude-haiku-4-5.

  Anthropic model benchmarked; showed high run-to-run variance, scoring 7 on one run and 13 on another

In every workspace you can read 18