ENTITY

UK AI Security Institute

Properties1
tagsai-security, research, data-poisoning, uk-aisi

1 min read

UK AI Security Institute

Type: organization (government AI-safety body) | Domain relevance: Co-author of the data-poisoning study behind the wiki’s “250 documents” correction.

Summary

The UK AI Security Institute (UK AISI) is a British government body focused on AI safety/security. Its Safeguards team co-conducted the “small samples poison LLMs” study with Anthropic and the Alan Turing Institute — the primary source the wiki uses to correct the 250 Authority Protocol (see Data Poisoning & LLM Backdoors). Its involvement is part of why the study is treated as credible research rather than vendor content.

Key Facts

  • UK government AI-safety/security institute.
  • Safeguards team co-authored the 250-documents poisoning study.
  • Lends the study cross-institutional credibility.

In the Sources

SourceContext
A small number of samples can poison LLMs of any sizeStudy co-author (with Anthropic + Alan Turing Institute).

Entities: Anthropic, Alan Turing Institute Concepts: Data Poisoning & LLM Backdoors, 250 Authority Protocol