Security scanning at Accenture used to take three to five business days. After deploying Cyber.AI, the company's new agentic security platform built around Anthropic's Claude, that same work now completes in under an hour. The company unveiled Cyber.AI on March 25, 2026, at RSA 2026 in San Francisco, framing it as the culmination of a multi-year partnership with Anthropic and placing Claude at the center of how one of the world's largest professional services firms handles enterprise cybersecurity.
The product is designed to move security operations from reactive, human-speed triage to continuous AI-driven monitoring. At its core, Claude serves as the reasoning layer, synthesizing data across the security stack, correlating alerts, and providing contextual guidance at each stage of the security lifecycle. Accenture combined Claude with its own library of proprietary security agents, built over two decades of cybersecurity delivery, to produce what the firm describes as an end-to-end agentic security capability.
Claude at the Core
The Cyber.AI architecture separates into three layers. The first is threat detection, where Claude ingests telemetry from across an enterprise's environment and identifies anomalies that would take a human analyst hours to surface. The second is response orchestration, where Claude coordinates remediation steps across teams and tooling, escalating where human judgment is needed and closing tickets where it is not. The third is governance, handled through a component called Agent Shield, which enforces organizational policies on the AI agents themselves, monitoring them in real time to ensure they operate within defined risk tolerances.
Accenture has already run Cyber.AI across its own global infrastructure: 1,600 applications and more than 500,000 APIs. The internal deployment produced the numbers the company is now using in its pitch to clients. Scan turnaround dropped from a range of three to five days to under one hour. Security testing coverage expanded from approximately 10 percent of the estate to over 80 percent. Those figures represent a change not just in speed but in the fundamental economics of enterprise security, where testing coverage has historically been constrained by the cost of skilled analyst time.
Cyber.AI by the Numbers
- Accenture internal apps secured1,600
- APIs covered500,000+
- Scan turnaround (before / after)3–5 days / under 1 hour
- Security testing coverage (before / after)~10% / 80%+
- Accenture professionals trained on Claude30,000
- Orgs citing AI vulnerabilities as fastest-growing risk87%
The Risk Landscape That Prompted the Build
The timing of the Cyber.AI launch tracks with a shift in the enterprise threat landscape that Accenture and others have been documenting for most of the past year. According to the World Economic Forum's Global Cyber Outlook Report 2026, which Accenture co-produced, roughly 87 percent of organizations now identify AI-related vulnerabilities as the fastest-growing category of cyber risk. That figure reflects two trends colliding: the rapid expansion of AI-generated code into production systems, bringing with it a new class of subtle logical flaws, and the rising use of AI by attackers to accelerate their own reconnaissance and exploitation pipelines.
For Accenture, the Cyber.AI build was also a direct consequence of its deepening work with Anthropic. The two companies announced a broader enterprise partnership in December 2025, with Accenture committing to certify roughly 30,000 of its professionals on Claude. Cyber.AI is the first major product to emerge from that relationship, and it represents a bet that the most durable value from the partnership will come not from deploying Claude in standard productivity workflows but from applying it to specialized, high-stakes operational domains where human capacity is genuinely the binding constraint.
"By combining Accenture's extensive library of proprietary agents with Claude, Cyber.AI builds on more than two decades of cybersecurity delivery expertise to help organizations move from human-speed response to continuous AI-driven cyber capabilities." Accenture, March 25, 2026
Governance as a Feature
One aspect of Cyber.AI that distinguishes it from earlier security automation products is the explicit focus on governing the AI agents themselves. Agent Shield, the governance component, is designed to address a concern that has grown alongside the adoption of agentic AI: the risk that autonomous agents acting on real infrastructure will drift outside the boundaries their operators intended. The component monitors agents in real time, audits their actions against organizational policy, and can intervene before a remediation step executes if it falls outside defined parameters.
That design choice reflects a tension Accenture and Anthropic are both navigating. The value of AI-driven security tools comes precisely from their ability to act without constant human sign-off. But in a security context, an agent that takes a wrong action, say, blocking legitimate traffic or quarantining a production system, can create the kind of outage it was meant to prevent. Agent Shield is Accenture's answer to that tension, though whether it holds in the edge cases that inevitably surface in production deployments is something only broader adoption will determine.
What This Signals for the Claude Security Stack
The Accenture announcement sits alongside a cluster of Claude-based security products that have emerged over the past several months. Claude Mythos, Anthropic's specialized cyber-offensive model, has been operating under restricted access through Project Glasswing, and Anthropic expanded its Claude Security product for enterprise codebase scanning into public beta in April. The pattern is a deliberate build-out of Claude's security credibility across the market stack: from the offensive research layer, which Mythos occupies, through the defensive enterprise layer, which Cyber.AI now anchors at the Accenture scale. Whether this consolidation around Claude holds as specialist security vendors develop their own model integrations will shape how competitive the security AI segment becomes by the end of the year.
