开源AI对齐诊断工具

meng shao (@shao__meng) 转发了 Sumanth (@Sumanth_077) 的帖子： Get a letter grade for your AI agent in under 5 minutes! iFixAi is an open-source diagnostic that catches AI operational misalignment before it damages your business. Actions your agent takes that don't match what you intended, designed, or expected. The dangerous part: these rarely show up in your usual KPIs. It runs up to 45 inspections against your agent in two tiers. 32 graded core inspections across five pillars - fabrication, manipulation, deception, unpredictability, and opacity. These produce the letter grade. 13 extended inspections across 11 new categories of frontier agent risk: sabotage, sandbagging, oversight evasion, power elevation, and more. These are exploratory - scored and reported separately without affecting the headline grade. They've run it against real open-source AI systems. OpenClaw, Hermes Agent, and Open WebUI all came back with an F. The results are published with full scorecards. Works with any provider. OpenAI, Anthropic, Gemini, Bedrock, Azure, HuggingFace, or any OpenAI-compatible endpoint. Plugs into CI so you can track alignment drift across every release. Every run writes a content-addressed manifest for bit-identical replay. Key capabilities: • 45 inspections: 32 graded core + 13 extended • Letter grade A-F from 5 pillars of misalignment risk • Extended tier covers sabotage, sandbagging, oversight evasion, power elevation • Provider-agnostic: OpenAI, Anthropic, Gemini, Bedrock, Azure, HuggingFace • Author custom fixtures for domain-specific testing • CI-friendly for tracking alignment drift over time • Content-addressed manifest for reproducible results 100% open source. I've shared the link in the replies!