1. Executive Summary (TL;DR)

2. Top IAM & Security News

1. Sophos State of Identity Security 2026 — 71% of orgs breached, $1.6M mean recovery

Published May 12. Vendor-agnostic survey of 5,000 IT and cybersecurity leaders across 17 countries and 14 industries. Headline numbers: 71% suffered at least one identity-related breach in the prior 12 months (Switzerland 89%, Mexico 83% leading by geography); average of three identity attacks per affected org; 67% of ransomware victims said the incident was directly tied to an identity attack; $1,637,363 mean / $750K median recovery cost; only 24% continuously monitor for unusual login attempts.

2. CVE-2026-41103 — Microsoft SSO plug-in (Jira/Confluence) authentication bypass, CVSS 9.1

Microsoft's May 12 Patch Tuesday disclosed an elevation-of-privilege flaw in the Microsoft Single-Sign-On Plug-in for Atlassian Jira & Confluence. An unauthenticated attacker can send a crafted response during the sign-in handshake to forge an identity and bypass Microsoft Entra ID authentication entirely, gaining read/write access to Jira and Confluence content. Rated "Exploitation More Likely" in Microsoft's Exploitability Index.

3. SailPoint Agentic Fabric — first dedicated identity-governance platform for AI agents

Announced May 11. SailPoint extends its Atlas/Identity Security Cloud foundation with a layer purpose-built for autonomous agents: agent discovery and registration, runtime policy enforcement, least-privilege scoping per task, automated credential rotation, and agent-aware threat detection. The platform integrates with major agent runtimes (LangChain, Bedrock AgentCore, Azure AI Foundry, Vertex Agent Builder) and exposes a unified inventory across human, machine, and agent identities.

4. CISA adds CVE-2026-42897 (Microsoft Exchange XSS) to KEV

On May 15 CISA added an actively-exploited Exchange Server cross-site scripting vulnerability to the Known Exploited Vulnerabilities catalog. Federal agencies were already racing a May 12 deadline for two earlier KEV adds — CVE-2026-32202 (Windows Shell, NTLMv2 hash leak residual from a partial February patch) and CVE-2024-1708 (ConnectWise ScreenConnect).

5. Instructure / Canvas — extortion agreement confirmed (May 12)

Instructure (parent of the Canvas LMS) publicly confirmed it had reached an "agreement" with the decentralized extortion group behind the breach disclosed the prior week. The campaign — attributed to UNC6040 / ShinyHunters lineage — threatened to publish stolen records from thousands of K-12 and higher-ed institutions, with reported initial access via vishing of help-desk staff to compromise an SSO account.

6. Microsoft Entra modernization: SCIM auth migration + passkeys for External ID GA

Microsoft's May 2026 What's New in Entra recap (covering changes landing this month) confirms passkeys for Entra External ID reach GA at the end of May, lets customers migrate SAP SuccessFactors provisioning from basic auth to workload identity federation in place (no reconfiguration required), and forces SCIM provisioning apps off OAuth 2.0 Authorization Code grant onto client-credentials or workload identity federation. Entra Global Secure Access now supports network-based file-type filtering for transfers to generative-AI and SaaS apps.

7. Quest research: NHI sprawl now outpaces every other identity discipline

Quest Security published Non-Human Identity Security in the Age of AI and Automation on May 14, restating what's becoming the consensus picture: NHIs (service accounts, API keys, agent credentials) are the fastest-growing identity class and the least-governed. The recommended modern stack: automated discovery, clear human ownership per credential, automated rotation, least-privilege scoping, and ITDR for behavioral baselines.

8. Grafana GitHub-token exposure (disclosed mid-window)

Grafana disclosed that an "unauthorized party" obtained a token granting access to its GitHub environment, with the source-code download attempted. Disclosure landed at the close of the reporting window; investigation and downstream-impact details are still developing.

3. AI, Identity & Emerging Tech

Ping Identity / KuppingerCole — From AI Agents to Trusted Digital Workers

Released this week. Core finding: AI agents are being deployed into production faster than enterprises can govern them, exposing gaps in identity systems originally designed for human users. The report calls out a specific failure mode — agents combining individually legitimate permissions in unintended ways, producing actions that bypass established controls and cannot be fully traced.

Cisco / CrowdStrike: 85% pilot, 5% production — the 80-point governance gap

Data referenced across multiple this-week reports from Cisco and CrowdStrike: 85% of enterprises are running AI-agent pilots, only ~5% have reached production. The blocker is governance, not capability — most agents lack distinct identities, durable audit trails, or behavioral baselines, and live in an "identity gray area" between human and machine.

Microsoft Defender for Cloud CIEM expands to AWS CloudTrail signals

Released this week. CIEM recommendations in Defender for Cloud now ingest AWS CloudTrail management-event activity alongside Access Advisor, and extend the inactivity lookback window from 45 to 90 days, evaluating unused role assignments rather than just sign-in activity.

IBM Vault 2.0 + unified Verify/Vault for the agentic era

Restated at industry events this week. IBM positions Vault 2.0 as identity-based security at scale for AI agents, with the unified Verify+Vault stack covering both human and non-human identities under one policy plane.

4. Cyber Threats & Attack Trends

a) "Log in, don't break in" — credential-led intrusion now dominates

Multiple this-week sources converge on the same point: stolen credentials, not unpatched vulns, are the dominant initial-access pathway. Sophos puts identity in 80%+ of ransomware operations; threat-intel firms tracked nearly 2.9B compromised credentials globally in 2025, with ~347M originating from infostealers on ~3.9M infected machines.

b) SaaS-extortion via help-desk vishing (ShinyHunters lineage continues)

The Instructure resolution this week is one bookend. The kill chain — voice phishing → help-desk-mediated SSO reset → OAuth scope abuse → mass SaaS export → leak-site extortion — remains the highest-yield identity-led attack pattern of 2026. Help-desk identity verification is the single point of failure.

c) Authentication-bypass surface area in identity middleware

CVE-2026-41103 in the Microsoft SSO plug-in for Jira/Confluence is the marquee example this week, but the broader pattern is identity-middleware (SSO plug-ins, federation bridges, SCIM connectors) becoming the soft underbelly between hardened IdPs and downstream SaaS. The flaw allowed an attacker to forge identity assertions and skip the Entra MFA challenge entirely.

d) Identity-edge appliances under sustained pressure

KEV additions over the trailing month — PAN-OS User-ID portal (CVE-2026-0300, prior week), Ivanti EPMM RCE, Cisco Catalyst SD-WAN auth bypass (CVE-2026-20127), BeyondTrust RS/PRA (CVE-2026-1731) — all hit appliances that enforce identity at the perimeter or manage privileged sessions. These boxes authenticate everyone else; their compromise is the highest-leverage identity-context takeover available.

e) NHI/secret leakage at the SDLC boundary

The Grafana disclosure (above) and ongoing infostealer harvesting both highlight that GitHub PATs, CI/CD secrets, and OAuth app credentials are now identity-class assets with prod blast radius — but few orgs govern them with the rigor applied to human SSO accounts.

5. Product Updates & Vendor News

6. Notable Research & Reports

Sophos — State of Identity Security 2026

Ping Identity / KuppingerCole — From AI Agents to Trusted Digital Workers

Quest Security — Non-Human Identity Security in the Age of AI and Automation (May 14)

Token Security — Ultimate Non-Human Identity Security Guide (May 13)

7. Practical Security Takeaways

  1. Patch CVE-2026-41103 immediately. Microsoft SSO plug-in for Jira/Confluence, CVSS 9.1, identity-forgery bypass. Then audit Jira/Confluence audit logs for anomalous SSO assertions over the past 60 days.
  2. Apply the May 12 KEV remediation set. CVE-2026-32202 (Windows Shell NTLMv2 leak — partial-patch problem), CVE-2024-1708 (ConnectWise ScreenConnect), and now CVE-2026-42897 (Exchange XSS, added May 15).
  3. Run a help-desk vishing exercise. Adversaries are walking through identity-reset workflows in production every week. Specifically validate that help-desk staff require out-of-band re-auth + manager approval before resetting MFA or password on any account with SaaS-admin or SSO-admin scope.
  4. Inventory non-human identities with named human owners. Pair every API key, service account, OAuth app, and agent credential with a responsible engineer. Credentials without owners get rotated or revoked. This single change is what makes the next four items work.
  5. Continuous identity monitoring, not quarterly review. Sophos found only 24% of orgs continuously monitor unusual logins. The fix is ITDR/UEBA-style behavioral baselines with real-time alerting on credential reuse, impossible travel, MFA-fatigue patterns, and OAuth-scope changes.
  6. Tighten SCIM and OAuth-app posture. Microsoft is forcing SCIM off Authorization Code grant for a reason — long-lived bearer tokens are the most-exploited identity-class secret outside of SSO sessions. Migrate to workload identity federation or client-credentials where the IdP supports it.
  7. Phishing-resistant MFA is now the floor for privileged accounts. Passkeys (FIDO2/WebAuthn) for admins, SaaS-admin, IdP-admin, and any role with cross-tenant scope. Number-matching push is no longer sufficient against MuddyWater-style live-coached MFA defeat.
  8. Treat AI agents as first-class identities. Per-task scoped credentials, short TTL, behavioral baselines on agent actions, and human-in-the-loop for any privilege-boundary crossing. If you can't answer "which agent did this, with what permission, on whose behalf" — fix that first.
  9. Rotate GitHub PATs, CI/CD secrets, and OAuth-app credentials older than 90 days. Grafana's disclosure is the third major incident in 2026 with GitHub-token blast radius. Bots and service identities tend to skip rotation policy entirely.
  10. Re-run AWS least-privilege reviews under the new 90-day window. Defender for Cloud's extended lookback will surface dormant identities and unused permissions you previously kept — clean them up before an attacker logs in instead.

8. Trends to Watch

9. Tool / Resource of the Week

**Token Security — *The Ultimate Non-Human Identity Security Guide*** (published May 13, 2026)

A practitioner-oriented playbook for inventorying, scoping, rotating, and monitoring NHIs across cloud, SaaS, containerized, and agentic stacks. Covers API keys, service accounts, OAuth tokens, secrets-manager hygiene, and agent credentials, with operational checklists and concrete owner-attribution patterns.

10. Sources