How we got frontier-level capability without a frontier dependency
How Box built an agentic, multi-model pen testing engine that proves logic flaws without a GPU farm.
AI Research
Invisible guardrails, infinite velocity: how modern CISOs secure autonomous agents without slowing innovation
How modern CISOs can secure autonomous AI agents, protect enterprise data, and enable innovation without sacrificing speed.
Why your content infrastructure matters more than ever
How secure, governed content infrastructure powers AI agents to operate, scale, and collaborate across the enterprise.
Rethinking what “secure” means in AI systems
Box CISO, Heather Ceylan, shares how AI systems demand a new definition of "secure" — one built around behavior, intent, and runtime control, not just access.
Gemini 3.7 Flash on real enterprise knowledge work
Gemini 3.7 Flash outperforms its predecessor on complex enterprise knowledge work, delivering higher accuracy and faster results.
We tested the models, not the room
Four AI sandbox escapes reveal that misconfigured containment, not model capability, is the real security risk we must fix.
Your AI agents will do whatever it takes to achieve their goal. Plan accordingly
Why AI agents will breach your systems and how to secure the data layer before they do.
Go from simply using AI to leading its rollout as a Box Certified AI Strategist
Introducing Box Certified AI Strategist: a free certification to help you lead, deploy, and govern AI across your organization.
Individual versus team: weighing the utility of just-for-you AI agents
Box's COO Olivia Nottebohm weighs the ROI of personal AI agents against scalable team agents for measurable business impact.
Claude Opus 5 on real enterprise work
Claude Opus 5 outperforms the prior generation on complex enterprise work, with major gains in exhaustive, multi-step analysis.
How government agencies should evaluate content management platforms in the age of AI
How government agencies can choose a content platform that meets compliance requirements and enables the safe adoption of AI.
The Hugging Face Intruder Was an OpenAI Model in an Eval
An OpenAI model autonomously hacked Hugging Face during a security test, revealing a new threat from unintentional, automated agents.
Gemini 3.5 Flash-Lite on real enterprise work
Box's evaluation finds Gemini 3.5 Flash-Lite is stronger, faster, and more efficient on complex, quantitative enterprise work.
Muse Spark 1.1 meets the mark on real enterprise work
Box's eval benchmark shows Meta Muse Spark competes with AI models on real enterprise knowledge work.
How GPT-5.6 handles real enterprise work
Box benchmarks GPT 5.6's Sol, Terra, and Luna models on real enterprise tasks across twelve industries.
Fable 5's guardrails have a cost, and we're the ones who pay
Box CISO, Heather Ceylan, explains how AI safety guardrails protect against attackers but quietly block the security teams defending your systems.
Serving the public, not the paperwork: How to free up time for HHS caseworkers
How AI and intelligent content management can free HHS caseworkers from paperwork and speed up benefits delivery.
A recap of Bio-IT World and DIA 2026
How Box helped life sciences organizations tackle dark data and build AI-ready content infrastructure at Bio-IT World and DIA 2026.
Claude Sonnet 5 raises the bar for enterprise intelligence across key industries
Claude Sonnet 5 delivers stronger enterprise document intelligence across key industries, built for production-scale AI deployment.
State of AI report: Overcome AI bottlenecks with a strong content foundation
How UK Northern Ireland and UK organisations can close the gap between AI adoption and real business value.