Box

Every IT leader knows the moment. Someone starts using ChatGPT for work. Then more people do. And suddenly you're governing something that was never designed for your enterprise in the first place. Jeff Chambers, VP of IT at WongDoody, has lived that moment. On the latest AI-First CIO episode, he shared three things every IT leader needs to hear: 1. Security breaks first. People's first experience with AI is personal and unsecured. When they carry that mindset into the enterprise, that's where the real challenge begins. 2. "Human in the lead" beats "human in the loop." Reviewing AI outputs isn't enough. Humans need to own the origin, evaluation, and ongoing vetting of every AI tool in the organization. 3. The ROI timeline is longer than you think. AI strategy isn't a project with an end date. It's ongoing change management — and the milestones matter as much as the technology. Jeff's framing that's stuck with us: "Processes need to be reevaluated from the ground up. Understand your complete business process across all of your systems, where your content is, and get that house in order." Watch the full episode: https://lnkd.in/g3tFpd2r What was the first thing that broke in your organization when AI started getting used at scale?

A 27-year-old bug in OpenBSD. A 16-year-old FFmpeg flaw. A FreeBSD vulnerability granting unauthenticated root access to anyone on the network. All found by AI. Overnight. Autonomously. Anthropic's Claude Mythos and Project Glasswing marks a genuine inflection point for enterprise security. The same AI that finds decades-old vulnerabilities can also help stop new ones from being created in the first place. Our CISO Heather Ceylan shares 4 things every security leader should be doing right now: 1. Revisit your threat models. 2. Get ahead of the Glasswing patch wave. 3. Talk to your vendors. 4. Invest harder in the fundamentals. "AI found a bug that the entire industry missed for 27 years. That should make every CISO uncomfortable and optimistic in equal measure. What matters now is what we do with the head start." Read Heather's full take here. 👇 https://lnkd.in/gj8KJfgj

There is a paradox hiding inside most enterprise AI strategies. The more agents you add, the harder it becomes to maintain consistency, trust, and control. It's not that individual agents fail. The problem is that when agents operate from different versions of the same document, different permission sets, and different content sources, their outputs inevitably diverge. More automation, but not more alignment. The fix isn't better orchestration. It's a better content foundation. When every agent works from the same governed, authoritative source of truth, outputs compound instead of conflict. That's the infrastructure decision that separates enterprises that successfully scale AI from those buried under their own agent sprawl. Read the full breakdown here. 👇 https://lnkd.in/gb3adFsC

Banks have embraced AI. Their primary hurdle has shifted to the burden of proof. With FINRA's 2026 Regulatory Oversight Report now published, the question regulators are asking isn't whether firms are using AI. It's whether they can prove how they used it. Forensic logs. Decision lineage. Documented supervision. The firms that can't show their work are the ones that will face the consequences. 3 steps separate firms building durable AI governance from those creating liability: ➡️ Proof over promise: Every AI-generated decision needs forensic logs and lineage. If you can't show your work, you can't defend your decisions. ➡️ Rationalize your stack: Fragmented AI experiments create tech debt and compliance blind spots. Every tool must serve a specific purpose tied to your strategy. ➡️ Keep humans in the loop: Agents acting without human validation are a liability. A two-way street of checks and balances is what makes governance defensible and revenue growth sustainable. The firms that prove their agents' homework won't just satisfy regulators. They'll be positioned to capture the revenue that manual, siloed processes are currently blocking. Full breakdown here. 👇 https://lnkd.in/gPxezGP9

This week in the Box Beat, we're covering everything happening at the intersection of AI and enterprise content. Here's what's inside: 👉 Claude Opus 4.7 on Box — 2x fewer model calls, 25% faster, 30% lower AI Unit usage 👉 Box MCP Apps now live in ChatGPT, M365 Copilot, and Glean 👉 Why enterprise search fails and how AI agents are fixing it 👉 For developers: automate workflows with the new Codex Box plugin and Claude Managed Agents Everything you need to stay ahead of enterprise AI this week. 👇

Box Extract was built to solve the problem most document processing pipelines hit in production. The prototype works. Then come the messy scans OCR can't handle, permissions that aren't enforced at query time, extracted data with nowhere structured to go, and defensive code written just to avoid API throttling errors. Box Extract handles all of it — not as a wrapper around a model, but as a compliant, secure, end-to-end extraction pipeline built for how enterprises actually work. The Box Extract Agent doesn't just pull text. It categorizes documents, enriches and validates extracted data against business rules, and routes uncertain results for human review. Once validated, data is saved as structured metadata attached directly to the source file — instantly searchable, filterable, and ready to trigger downstream workflows in Box Relay, power dashboards in Box Apps, or feed external systems like Salesforce and Snowflake. No operational debt. No prompt engineering. No data leaving your security perimeter. See what's under the hood. 👇 https://lnkd.in/gXsk2Fps

Most enterprise search is broken. And the fix starts with your file system. In our latest episode of AI Explainer, our CTO Ben Kus breaks down why agentic search and context retrieval are transforming how enterprises find and use unstructured data, and why the file system underneath it all is what makes it possible. Instead of returning a list of links for humans to click through, AI agents can now navigate your file system like a person, reading what's inside each file and surfacing exactly the context needed to get work done. "Agentic search might be one of the most critical ways that search has evolved in enterprises, especially on unstructured data, so that people can actually find what they're looking for." Watch here. 👇 https://lnkd.in/gM7EgpQm

"AI transformation is still human-led change." Our COO Olivia Nottebohm shares the unfiltered story of Box's year-long AI journey, starting with 97 sandbox agents and no scalable strategy, to building a four-phase playbook, a company-wide AI certification program, and a network of AI managers across every subfunction. The technology evolved. The strategy shifted. The lessons came from doing the work, not planning it. If you're leading an organization through this right now, Olivia's perspective is one of the most honest and practical breakdowns of what becoming AI-first actually looks like. https://lnkd.in/gj2KWMpj What's the hardest part of your AI transformation so far? 👇

Anthropic just released Claude Opus 4.7 — and Box has the benchmark data to show what it means for enterprise AI. The results challenge a common assumption — that efficiency and quality are a tradeoff. Opus 4.7 didn't just get faster. It did so while delivering the same quality of output. In agentic AI, every extra step has a cost: more latency, more complexity, more spend. The best models don't just produce great answers — they find the shortest path to them. Compared to Opus 4.6, Opus 4.7 delivered: • 56% fewer model calls • 50% fewer tool calls • 25% faster task completion • 30% fewer tokens used The gains were in how Opus 4.7 reached its answers, not in what it produced. Since the Box Agent consumes AI Units based on tokens used, Opus 4.7's reduction in token usage translates directly to lower AI Unit consumption — meaning more work, for less. For enterprises building at scale, that's not just a performance improvement — it's a change in what's practical to deploy. Claude Opus 4.7 will be available today in Box AI Studio and through the Box API. Read the full eval: https://lnkd.in/gRzmRUhb

Every enterprise AI conversation eventually hits the same wall: you can build the agent, but you can't give it governed access to the content it needs. Box + IBM are solving that. Secure content + AI orchestration = agentic workflows that actually work in regulated, complex environments. Don't miss IBM's live session on the interoperability blueprint this Thursday. Worth your time if you're building (or buying) in this space. Register here: https://lnkd.in/gbA7rR77