• Dotika
  • Posts
  • Risk rubric democratizes AI safety

Risk rubric democratizes AI safety

ALSO : AlterEgo: telepathic communication is here

Author

Miso PECIREP
September 18, 2025

News of the day

1. RiskRubric.ai launches to standardize AI model risk assessment, offering transparent A-F grades across six pillars. It empowers informed decisions and builds trust in open-source AI. → Read more

2. AlterEgo enables telepathic thought capture and communication, offering instant idea recording, device control, and real-time translation. It enhances productivity and accessibility, marking a new era in human-computer interaction. → Read more

3. IBM launches Granite-Docling-258M, an open-source document AI model. It excels at layout-faithful extraction, preserving tables, code, and equations for enterprise use. → Read more

4. Gemma 3 introduces multimodal capabilities, allowing it to process both text and images simultaneously. This enables advanced tasks like image description, data analysis from charts, and multilingual understanding, pushing AI interaction forward. → Read more

Our take

Hi Dotikers!

Your board speaks in letters, your data scientists in tokens, RiskRubric acts as interpreter. The novelty is simple yet powerful: a report card from A to F across six pillars – transparency, reliability, security, privacy, societal safety, and reputation – based on hundreds of robustness tests and adversarial attacks. Finally, a comparable score that ends the “trust me” era and ushers in the “show me” era.

Our opinion is straightforward: adopt it as a product safeguard. RiskRubric already publishes monthly dashboards on more than 150 models, backed by an independent consortium. It is a market-ready building block that clarifies trade-offs between performance and risk.

On the governance side, the chain is becoming credible. The NIST AI RMF provides a structured risk management process, and ISO/IEC 42001 lays the foundation for AI management systems. By integrating RiskRubric scores with these frameworks, you create a common language from the audit committee to the ML team, with thresholds, evidence, and action plans.

On the compliance side, Europe will not settle for promises but will demand verifiable elements. The AI Act is moving forward, with obligations for GPAI models starting in 2025 and broader enforcement in 2026. An external, documented score reduces approval friction and secures your due diligence.

The roadmap is clear: set an entry threshold (for example a global B or 75 per pillar), treat gaps as dated remediation plans, and feed production monitoring with detected vulnerabilities. Nothing exotic, just disciplined tooling.

Tokens talk, letters cut through.

M.

Tweet of the day

Reply

or to participate.