AI Security Institute
The AI Security Institute (AISI) is a research organisation under the Department for Science, Innovation and Technology, UK, that aims "to equip governments with a scientific understanding of the risks posed by advanced AI". It conducts research and develop and test mitigations. Previously, it was k
Signal Radar
Five-axis snapshot of this entity's footprint
Mentions × Lab Attention
Weekly mentions (solid) and average article relevance (dotted)
Timeline
1- Research MilestoneJul 3, 2026
AISecInst finds test-time compute budgets skew frontier model evaluations
View source
Relationships
2Uses
Regulated
Recent Articles
2AISI: Fixed compute budgets underestimate AI agents by 60%
~AISI found standard benchmarks cap compute budgets, underestimating agent capabilities by ~60%. Success rates jumped ~25% with 10x tokens.
96 relevanceAI Security Inst Shows Test-Time Compute Skews Frontier Evaluations
~AISecInst research shows test-time compute budgets skew frontier model evaluations, challenging standard practices.
92 relevance
Predictions
No predictions linked to this entity.
AI Discoveries
2- hypothesisactiveJul 3, 2026
H: Within 60 days, a major AI lab (OpenAI, Google DeepMind, or Anthropic) will publish a paper or blog
Within 60 days, a major AI lab (OpenAI, Google DeepMind, or Anthropic) will publish a paper or blog post acknowledging the test-time compute evaluation vulnerability and proposing a standardized benchmark correction methodology.
70% confidence - observationactiveJul 3, 2026
Research: Test-Time Compute & Evaluation Integrity [accelerating]
State of art: AISecInst research shows test-time compute budgets systematically skew frontier model evaluations, challenging the validity of current benchmark comparisons.. Key insight: If test-time compute is unaccounted for, reported benchmark gains may reflect inference strategy rather than genui
70% confidence
Sentiment History
| Week | Avg Sentiment | Mentions |
|---|---|---|
| 2026-W27 | 0.15 | 2 |