Blog

Security Insights

Deep dives, expert analysis, and practical guidance on exposure management, adversarial validation, and the future of AI-driven exposure management.

Running DeepSeek V4.1 Flash in production: serving, KV caching and autoscaling on AWS
engineeringAI Security

Running DeepSeek V4.1 Flash in production: serving, KV caching and autoscaling on AWS

A technical guide to self-hosting DeepSeek V4.1 Flash for long-context agent workloads: vLLM configuration, a latency model for sizing, tiered and shared KV cache in S3, prefix-aware routing, autoscaling and unit costs.

Nov 6, 202613 min
Tooling as Code and Structured Scratchpads
engineeringOffensive Security

Tooling as Code and Structured Scratchpads

The two changes that moved our AI security agents furthest this year had nothing to do with the model. Here is how tooling as code and structured scratchpads reshaped how the agents act, and how they prove a finding is real.

Aug 3, 202615 min
How to Write an Effective AI Agent Skill Four-Layer Architecture
engineeringOffensive Security

How to Write an Effective AI Agent Skill: The Four-Layer Architecture

Most teams building AI agents get the ratio wrong: 90% code, 10% methodology. Here is the four-layer architecture Strobes uses to build skills that run complete security assessments autonomously.

Mar 31, 20267 min
AI Harness for Offensive Security - Strobes blog cover showing multi-agent architecture concept
Offensive SecurityPenetration Testing

Building an AI Harness for Offensive Security: What It Takes to Turn LLMs Into Reliable Pentest and Validation Operators

The model is 20% of the problem. Here is the engineering story behind the orchestration, tooling, middleware, and infrastructure that turns a capable LLM into a reliable penetration testing operator.

Mar 22, 202614 min
Cut RDS Costs by 50% with Aurora Serverless V2 Idle Connection Fix
engineering

Cut RDS Costs by 50% with Aurora Serverless V2 Idle Connection Fix

In a recent migration from a standard RDS DB instance to Aurora RDS PostgreSQL Serverless V2, we encountered an unexpected issue a significant and unexplained increase in Aurora Serverless V2 connections. This anomaly led to spikes in resource utilization, specifically memory, and caused a noticeabl

Oct 25, 20244 min