Deeper Than Passwords: The Passkey Boom and FIDO in the Global Authentication Race

A new European pre-standard, prEN 18286, offers a structured quality management system path to EU AI Act compliance that complements ISO 42001. With the August 2026 high-risk enforcement deadline approaching, organizations need to understand where the two standards converge and how to prepare.

The FIDO Alliance has launched two dedicated working groups for agentic authentication and payments, with Google donating its Agent Payments Protocol (AP2) and Mastercard contributing Verifiable Intent. The move consolidates previously independent efforts into an open, industry-governed standards process backed by major payment networks, credential managers, and AI labs.

The Cloud Security Alliance's nonprofit arm, CSAI Foundation, announced a series of milestones at its Agentic AI Security Summit: a new Catastrophic Risk Annex translating extreme AI scenarios into auditable controls, registration as a CVE Numbering Authority for AI vulnerabilities, and an empirical finding that repeated adversarial testing can cause agents to refuse legitimate tasks.

Microsoft's April 2026 Patch Tuesday addresses 164 vulnerabilities - double March's count - including one actively exploited SharePoint zero-day, a disclosed Defender privilege escalation linked to the BlueHammer exploit, and a CVSS 9.8 unauthenticated RCE in the Windows IKE Extension. Browser updates from Google and Mozilla add another layer of urgency.

An AI-powered audit uncovered 38 vulnerabilities in OpenEMR - the open-source EHR platform serving over 200 million patients - while Medtronic confirmed a breach by ShinyHunters involving an alleged 9 million stolen records. Together, these events underscore the healthcare sector's persistent security challenges across both software supply chains and corporate IT infrastructure.

The Cloud Security Alliance introduced its AI Security Maturity Model (AISMM) at RSAC 2026, designed to give enterprises measurable KPIs for AI security across categories including model security, agentic applications, and MCP servers. Paired with a restructured Enterprise Membership program, CSA is betting that scalable analyst support - not just published research - is what organizations need to close the governance gap.

The Cloud Security Alliance has published a structured decision framework to help security and risk teams select the right AI governance standard. The guide distills the choice down to seven questions spanning jurisdiction, risk profile, governance maturity, and supply chain position.

A critical sandbox escape flaw (CVE-2026-5752, CVSS 9.3) in Cohere AI's Terrarium Python sandbox allows attackers to execute arbitrary commands as root via JavaScript prototype chain traversal. The project is abandoned, so no patch is forthcoming.

The UK's National Cyber Security Centre has formally recommended passkeys as the default authentication method, calling passwords "no longer resilient enough for the contemporary world." The announcement at CYBERUK 2026 coincides with Microsoft beginning its rollout of Entra passkey support to Windows devices, including unmanaged PCs.

Two emerging open standards - OpenEoX and CLE - are joining forces to give organizations machine-readable visibility into whether the software and hardware they depend on is still supported. Their collaboration addresses a critical blind spot in supply chain security, just as the EU Cyber Resilience Act makes lifecycle transparency a legal requirement.

Enterprise backup strategies are falling short in ransomware scenarios, with backup usage for recovery hitting a six-year low. The real risk isn't data loss-it's prolonged operational downtime that backups alone cannot prevent.

Identity weaknesses now appear in nearly 90% of breach investigations, and new CSA research confirms most organizations still cannot distinguish AI agent actions from human ones. As both sides of the adversary-defender equation deploy autonomous agents, the identity layer has become the decisive battleground - and the industry's response is racing to catch up.

Three Cloud Security Alliance research efforts released in the same week reveal a unified problem: AI agents are operating at enterprise scale with borrowed identities, excessive permissions, and invisible trust chains - and most organizations lack the controls to detect or contain the resulting risks.

A Lumma Stealer infection at AI vendor Context.ai cascaded through an overly permissive OAuth connection into Vercel's enterprise Google Workspace, exposing customer credentials and internal data. The incident - now claimed by a threat actor using the ShinyHunters name - is a textbook case of the supply chain risks that AI integrations introduce.

A critical code injection flaw in protobuf.js (CVE-2026-41242, CVSS 9.4), downloaded over 52 million times per week, allows attackers to achieve remote code execution by crafting malicious protobuf schema definitions. Public proof-of-concept exploit code is now available, and patches have been released in versions 8.0.1 and 7.5.5.

While passkey adoption has crossed the mainstream threshold, the harder engineering work is now underway: portable credentials, standardized wallet certification, and cryptographic delegation chains for AI agent transactions. A wave of specification updates and new board appointments signals the ecosystem is solving the interoperability and governance problems that will determine whether passwordless authentication scales durably.

With 4 billion passkeys in active use, 87% enterprise adoption, and Microsoft auto-migrating tenants to passkey profiles, the convergence of data in April 2026 suggests password-based authentication is entering terminal decline - though the transition is far from complete.

New CSA survey data reveals 53% of organizations have experienced AI agent scope violations, while research into shared workspace authorization failures and a new runtime enforcement product from Curity illustrate both the depth of the problem and the emerging architectural responses.