Almanac
technique

Policy Vulnerability Testing

techniqueactiveprovisionalpolicy-vulnerability-testing-7d44d961·2 events·first seen 13d ago

Aliases: Policy Vulnerability Testing

Co-occurring entities

More like this (12)

Recent events (2)

5Anthropic News·13d ago·source ↗

Anthropic publishes elections-risk testing methodology and releases automated evaluation tools

Anthropic describes its two-stage process for identifying and mitigating elections-related risks in Claude: qualitative 'Policy Vulnerability Testing' (PVT) conducted with external subject matter experts, followed by large-scale automated evaluations. The post details how findings from PVT inform mitigation strategies such as policy updates, model fine-tuning, and response behavior changes, with a case study on election administration accuracy. Anthropic is also releasing some of its automated evaluation tools publicly to help other organizations improve election integrity efforts.

6Anthropic News·13d ago·source ↗

Anthropic details red teaming methods and calls for standardized AI testing practices

Anthropic published a detailed overview of red teaming approaches used to test Claude and other AI systems, covering domain-specific expert testing, automated red teaming, multilingual/multicultural testing, and multimodal red teaming. The post documents empirical findings about when each method is appropriate, highlights partnerships with organizations like Thorn, Institute for Strategic Dialogue, and Singapore's IMDA, and closes with policy recommendations for building a standardized AI testing ecosystem. The piece is notable for its operational specificity and its explicit call for industry-wide standards to enable cross-system safety comparisons.