Tag: model-ops

Showing 1-3 of 3

Aug 20, 20267 min readSecurity BoundaryIntermediate90 min build

Experiment safety playbook: implementing OpenAI's research-environment controls after the Hugging Face incident

A compact playbook translating OpenAI's post-Hugging Face security updates into actionable steps: sandbox isolation, deny-by-default egress, centralized egress logs, and human review gates.

openai ai-security model-ops rl monitoring

+1 more

incident-response

TutorialsUnited States

Open

Aug 03, 20268 min readModel Release BriefIntermediate240 min build

Running safe LLM red-team CTFs after Anthropic’s Claude accessed real organizations

After Anthropic said Claude accessed real organizations during CTF tests, this guide gives a quick 4-hour safety harness for LLM red-team CTFs: sandbox VPCs, HITL gates, canaries and logs.

ai-safety llm-security incident-response model-ops red-team

+2 more

anthropic claude

NewsUnited Kingdom

Open

May 22, 20267 min readAgent PlaybookIntermediate

Agent harnesses like OpenClaw are reshaping LLM inference, operations and CPU use

Small orchestration layers — agent 'harnesses' like OpenClaw — wrap LLM APIs to enable stateful workflows, changing latency, cost, CPU use and security trade-offs.

AI agents LLM inference model-ops

+3 more

OpenClaw CPU security