Barracuda signs OpenAI’s call for collective action on cyber defense

A CISO’s perspective on why accelerating AI‑powered defense is now essential as vulnerabilities surge and attacker capability rapidly advances.

Key takeaways

• Barracuda signed because defenders need capable AI faster than attackers will get it.  Accelerating AI‑powered defense is essential for customers, MSPs and partners facing rising threats and limited resources.
• Published vulnerabilities are doubling while security headcount is not, creating a widening gap that requires automation and scalable controls rather than more manual work.
• Collective action across industry, government and frontier labs is now required

As Barracuda’s Chief Information Security Officer (CISO), I’m sharing my perspective on why we chose to sign OpenAI’s call for collective action on cyber defense. This commitment reflects the realities our customers, MSPs and channel partners face every day — and the urgent need to accelerate AI‑powered defense across organizations of all sizes.

Why Barracuda signed

OpenAI’s call aligns directly with the challenges facing the organizations we protect. Published vulnerabilities are doubling year over year, while security headcount is not. Many customers already make difficult decisions about which findings to leave open. MSPs and channel partners absorb this across every client simultaneously — a provider supporting 40 organizations manages 40 backlogs growing at the same accelerating rate.

When the input doubles and capacity does not, something has to give. AI‑powered defense is one of the few scalable ways to change that equation.

The statement’s central claim is that defenders have a window in which capable AI models can help fix weaknesses that have “accumulated for years,” and that window narrows as attacker capability catches up. My research supports this assertion, demonstrating that the backlog of vulnerabilities will inevitably be uncovered in the coming years. The main question is how many will be detected and addressed by defenders before attackers find them.

The arithmetic of the challenge

Published vulnerabilities climbed from just under 40,000 in 2024 to roughly 48,000 in 2025. Based on observed data through August 2026, we expect to see approximately 88,000 this year — with the second half running even faster. Whatever else is true about AI and security, this is the number every proposal must survive.

Resource-constrained organizations are struggling to keep up. The question is no longer how to hire more people — it is which parts of the work can be done accurately with AI.

This is the reality our customers and MSP partners live with every day.

Published CVEs by year, excluding rejected records. 2026 is 57,806 observed through 2026-08-31 across 241 publishing days, annualized on a publishing-days divisor. The last six months ran at a higher rate than that, so the projection is the conservative reading. Source: NVD. Snapshot 2026-08-31.

What needs to be added to the call for collective action

The call for collective action is an important start — and I support its core principles. But from a defender’s perspective, it does not yet go far enough. To close the widening gap between vulnerability discovery and remediation, we need additional commitments from governments, frontier AI companies and software providers. The following additions would make the call materially more effective for the organizations on the front lines.

Governments: Pay for the change of state, not the headcount
A time limited tax incentive tied to demonstrated security improvements — not hiring alone — would meaningfully accelerate progress. Such an incentive could be tiered based on where an organization starts:

• External attack surface inventory: produce a verified inventory and move all management interfaces off the public internet.
• Phishing resistant authentication: enforce it for all administrative access.
• Patch latency measurement: measure and publish patch latency for internet facing systems.

Qualify roles by remediation, not the word “security”
General security hiring incentives tend to fund “finders,” not fixers. Assessments end in deliverables, not in removal of weaknesses. Remediation requires engineering access, system ownership and change windows.

Incentives should therefore target:

• Remediation of vulnerabilities in infrastructure and code
• Automation that performs remediation
• Tools and services that close findings for organizations too lean to hire remediation engineers

This also closes a loop left open in the call: automated discovery without funded remediation capacity simply moves work downstream. Both ends of the pipeline must be supported.

Given global cybercrime costs estimated around US $500 billion, such incentives may well pay for themselves.

Frontier AI companies: Build the missing capability

Automated vulnerability discovery and triage — with triage the harder half — is the clearest capability frontier labs can deliver across LLM-based code scanning, dynamic application security testing and autonomous red-teaming.

Discovery attracts attention; triage eliminates weaknesses. A system surfacing 10,000 findings for a project maintained by two volunteers shifts burden onto those least able to absorb it. The full path must be funded:

• Find the weakness
• Determine whether it is reachable in real configurations
• Rank it
• Produce a candidate fix
• Verify the fix holds

This is the capability that makes “authorized testing, private disclosure and verified fixes” possible at scale.

Software companies: Include essential security features

One more principle applies to software providers: security functionality should be included, not upsold. Charging separately for audit logs or single sign‑on (SSO) turns essential security features into revenue lines, and the organizations that decline upgrades are often the least able to absorb an incident.

Author: Arve Kjoelen. Source: OpenAI, “A call for collective action on cyber defense”, openai.com/collective-cyberdefense , retrieved 31 Aug 2026. All quotations are from that page.

[1] “Global Cybercrime Damages: A Baseline for Frontier AI Risk Assessment”.  https://arxiv.org/abs/2603.20570.  Retrieved 31 Aug 2026.