Reinforced Labs ResearchResearch notes and explainers on AI evaluation, guardrails and agent security.https://reinforcedlabs.ai/research2026-10-27T12:00:00ZReinforced Labs ResearchAI compliance testing: EU AI Act, NIST AI RMF and ISO/IEC 42001https://reinforcedlabs.ai/learn/ai-compliance-testing2026-10-27T12:00:00Z2026-10-27T12:00:00ZWhat each framework actually asks you to test, mapped to concrete evaluations and evidence.LLM-as-a-judge: how it works, where it fails, how to validate ithttps://reinforcedlabs.ai/learn/llm-as-a-judge2026-10-24T12:00:00Z2026-10-24T12:00:00ZJudge designs, documented biases, and a validation protocol with agreement metrics.How many test cases does an LLM eval need?https://reinforcedlabs.ai/research/eval-sample-size2026-10-21T12:00:00Z2026-10-21T12:00:00ZWilson intervals, power analysis and a Monte Carlo study of paired vs. unpaired tests. Detecting a 2-point regression takes thousands of cases, not hundreds.EU AI Act testing requirements and the 2026–2028 timelinehttps://reinforcedlabs.ai/learn/eu-ai-act-testing-requirements2026-10-20T12:00:00Z2026-10-20T12:00:00ZArticles 9, 15 and 55, the Digital Omnibus delay, and what evidence a test programme should produce.LLM-as-a-judge vs. human evaluationhttps://reinforcedlabs.ai/compare/llm-as-a-judge-vs-human-evaluation2026-10-17T12:00:00Z2026-10-17T12:00:00ZCost, agreement, latency and failure modes compared, and a hybrid design that uses each where it is strongest.AI safety and evaluation for financial serviceshttps://reinforcedlabs.ai/industries/financial-services2026-10-16T12:00:00Z2026-10-16T12:00:00ZModel risk management (SR 11-7), credit and fraud use cases, and agent controls for payments.Your LLM judge is a diagnostic test: correcting measured failure rateshttps://reinforcedlabs.ai/research/llm-judge-error2026-10-14T12:00:00Z2026-10-14T12:00:00ZSensitivity, specificity and the Rogan–Gladen correction. At a 0.5% true failure rate, a 97%-accurate judge reports 3.4%.NIST AI RMF Measure for generative AI: an evaluation checklisthttps://reinforcedlabs.ai/learn/nist-ai-rmf-measure2026-10-13T12:00:00Z2026-10-13T12:00:00ZTurning the Measure function and the NIST AI 600-1 profile into tests you can run.Indirect prompt injection, explainedhttps://reinforcedlabs.ai/learn/indirect-prompt-injection2026-10-10T12:00:00Z2026-10-10T12:00:00ZHow instructions hidden in documents, web pages and tool output hijack agents, with real research and the controls that hold.LLM evaluation plan templatehttps://reinforcedlabs.ai/templates/llm-eval-plan2026-10-09T12:00:00Z2026-10-09T12:00:00ZA copy-ready plan: scope, threat model, categories, sample sizes, graders, gates and owners.pass@k vs. pass^k: measuring agent reliabilityhttps://reinforcedlabs.ai/research/pass-at-k-vs-pass-hat-k2026-10-07T12:00:00Z2026-10-07T12:00:00ZThree task populations with identical pass@1 = 0.60 have pass^8 between 1.7% and 28.8%. How to estimate pass^k and how many trials you need.Evaluating clinical and healthcare LLM applicationshttps://reinforcedlabs.ai/industries/healthcare2026-10-06T12:00:00Z2026-10-06T12:00:00ZClinical harm taxonomies, expert graders, over-refusal and documentation for regulated deployments.Prompt injection vs. jailbreak: what's the difference?https://reinforcedlabs.ai/compare/prompt-injection-vs-jailbreak2026-10-03T12:00:00Z2026-10-03T12:00:00ZDifferent attacker, different target, different defences. A precise comparison with examples.AI red-team report templatehttps://reinforcedlabs.ai/templates/ai-red-team-report2026-10-02T12:00:00Z2026-10-02T12:00:00ZFinding format, severity model, reproduction rate and retest log, ready to copy.A 2% attack success rate is not low: retry economics of prompt injectionhttps://reinforcedlabs.ai/research/prompt-injection-retry-risk2026-09-30T12:00:00Z2026-09-30T12:00:00ZCumulative compromise probability, and why correlated defence layers deliver 3–30× less protection than the independence assumption predicts.OWASP Top 10 for LLM Applications (2025), explained for engineershttps://reinforcedlabs.ai/learn/owasp-top-10-llm2026-09-26T12:00:00Z2026-09-26T12:00:00ZAll ten risks, what each looks like in an agent, and the test and control that addresses it.Guardrail thresholds under low base rateshttps://reinforcedlabs.ai/research/guardrail-thresholds2026-09-23T12:00:00Z2026-09-23T12:00:00ZWith a 0.1% attack rate, a classifier with AUC 0.96 tuned to 95% recall has 0.5% precision. Cost-optimal thresholds, worked through.MCP security: tool poisoning, rug pulls and how to defendhttps://reinforcedlabs.ai/learn/mcp-security2026-09-19T12:00:00Z2026-09-19T12:00:00ZThe Model Context Protocol attack surface and a hardening checklist for teams connecting agents to MCP servers.The latency budget of AI guardrails: sequential, parallel and hedgedhttps://reinforcedlabs.ai/research/guardrail-latency2026-09-16T12:00:00Z2026-09-16T12:00:00ZOrder statistics of parallel checks. Hedging at the single-check p95 cuts eight-check p99 from 49 ms to 35 ms for 5% extra load.Evals vs. guardrails vs. red teaminghttps://reinforcedlabs.ai/compare/evals-vs-guardrails2026-09-15T12:00:00Z2026-09-15T12:00:00ZComparison.