Tag: model-risk

Showing 1-3 of 3

Aug 24, 20267 min readSecurity BoundaryIntermediate120 min build

Create auditable model-risk tests from FelonyBench's felonies.json for CI pipelines

Fetch and archive FelonyBench's felonies.json, pick high‑impact labels (e.g., production database compromise 4×), then run sanitized CI tests to produce auditable model-risk reports.

felonybench ai-cybersecurity benchmark model-risk ci

+1 more

auditability

NewsUnited Kingdom

Open

Jun 23, 20267 min readModel Release BriefIntermediate

Outsourced annotators report using public chatbots to generate 'human' training dialogues

Whistleblowers say subcontracted annotators often prompt public chatbots (e.g. ChatGPT) to produce 'human' training dialogues—risking model feedback loops. Practical checklist included.

training-data data-quality crowdwork model-risk ai-safety

+2 more

data-governance newscientist

NewsUnited States

Open

Feb 05, 20268 min readModel Release BriefIntermediate5 min build

Super Bowl LX: Platform-branded AI ads, creative risks, and builder priorities

Super Bowl LX may spotlight platform-branded AI ads — from Anthropic’s jab at OpenAI to Google’s Gemini flub. This brief outlines risks and practical safeguards for teams.

super-bowl ai-ads advertising anthropic openai

+3 more

google-gemini marketing model-risk