A critical pre-authentication RCE in OpenJDK is the story every Java shop needs to hear today. CVE-2026-114938 - already being called Java4Shell - affects Jira, Jenkins, Keycloak, Confluence, and any application accepting untrusted Base64 input on modern x86-64 hardware. CISA published an urgent advisory, and Talos released same-day detection guidance. Meanwhile, a supply-chain attack through Optimizely’s A/B testing platform served malware to visitors of 178 major websites, and Russian APT Cold Relic demonstrated how one cloud breach becomes two when incident response is scoped too narrowly.
In the News
Java4Shell CVE-2026-114938 - Pre-Auth RCE Across Enterprise Java
The vulnerability is in OpenJDK’s Base64 decoder, specifically in the AVX-512 VBMI optimized code path introduced in JDK 17.0.6. A malformed Base64 input triggers a buffer overflow in the SIMD processing logic, allowing an unauthenticated attacker to achieve remote code execution on any Java application that processes untrusted Base64 data - which includes authentication endpoints, REST API handlers, and file upload mechanisms.
The practical blast radius is enormous. Jira, Jenkins, Keycloak, and Confluence all accept Base64 input at pre-authentication endpoints. Any organization running these applications on Intel Xeon Scalable (Ice Lake or later) or AMD EPYC (Genoa or later) processors is affected - and that covers the majority of cloud and enterprise server hardware deployed in the last four years. The vulnerability was discovered by an AI-assisted fuzzer, which found the SIMD edge case that human auditors had missed across multiple code reviews.
CISA’s advisory urges immediate patching to OpenJDK 17.0.14 or 21.0.6, both of which contain the fix. No exploitation has been observed in the wild yet, but the combination of a pre-auth attack vector, a detailed public advisory, and a straightforward trigger mechanism means weaponization timelines will be measured in days, not weeks.
What defenders should do: Patch OpenJDK immediately. If patching requires a maintenance window, deploy WAF rules to validate and reject malformed Base64 input at the application boundary. Inventory all Java applications and confirm their JDK version - many organizations have lost track of which JDK version runs where. MITRE ATT&CK: T1190 (Exploit Public-Facing Application).
Optimizely Supply-Chain Attack Hits 178 Websites Including CNN, BBC, Zoom
Attackers compromised Optimizely’s experimentation platform and used it to deliver multi-stage malware to visitors of at least 178 high-traffic websites. The hijacked A/B testing scripts - loaded by sites as trusted first-party resources - deployed card skimmers on e-commerce pages, credential stealers on login forms, and crypto-swappers that replaced cryptocurrency wallet addresses on financial pages.
The attack is particularly effective because A/B testing scripts are designed to modify page content dynamically. Content Security Policy headers on most affected sites whitelisted Optimizely’s domain, giving the malicious payloads full access to the DOM without triggering browser security controls. Report URI, the security monitoring service that first identified the anomalous script behavior, confirmed that the injected code was served selectively - using Optimizely’s own targeting logic to deliver different payloads to different visitor segments, complicating detection.
What defenders should do: Audit all third-party JavaScript loaded on your web properties. Implement Subresource Integrity (SRI) hashes for external scripts where the vendor supports it. Deploy Content Security Policy with strict source directives and monitor for violations. For organizations whose employees visited affected sites, check endpoint telemetry for credential stealer indicators. MITRE ATT&CK: T1195.002 (Supply Chain Compromise: Compromise Software Supply Chain).
Cold Relic Chains Google Workspace Breach into Second U.S. Organization
Google Mandiant disclosed that Cold Relic, a Russian state-backed espionage group, exploited data harvested from a compromised Google Workspace tenant at one U.S. organization to successfully breach a second, unrelated U.S. organization. The attackers used contacts, calendar entries, and shared document links from the first victim to craft highly convincing spearphishing messages that bypassed the second organization’s email security controls.
The chain illustrates a systemic gap in incident response scoping. The first organization’s IR team contained the breach within their own tenant but did not notify partner organizations that may have been exposed through shared data. Cold Relic exploited that gap - using the trust relationship between the two organizations as an implicit credential. This pattern of inter-organizational lateral movement is particularly dangerous in environments with federated identity, shared cloud workspaces, or joint project repositories.
What defenders should do: Expand IR playbooks to include downstream partner notification and token revocation for any organization that shared data with the compromised tenant. Deploy cross-tenant anomaly detection for OAuth tokens and API calls. Enforce step-up authentication for sessions originating from unfamiliar devices or locations. MITRE ATT&CK: T1199 (Trusted Relationship), T1566.001 (Spearphishing Attachment).
Puppet Server RCE CVE-2026-115731 - CISA KEV Addition
CISA added CVE-2026-115731 to the Known Exploited Vulnerabilities catalog after confirming active exploitation of Perforce Puppet Server. The flaw allows any client with a CA-signed certificate - including any managed node in a Puppet deployment - to execute arbitrary code on the Puppet server. A single compromised endpoint in a Puppet-managed fleet becomes a direct path to infrastructure-wide control.
What defenders should do: Patch Puppet Server immediately. Audit client certificate inventory and revoke certificates for decommissioned or unnecessary nodes. Segment the Puppet server so only authorized management networks can reach it. Treat the Puppet server as a Tier 0 asset - equivalent to a domain controller in terms of security criticality.
Defender Action Items
- Patch OpenJDK to 17.0.14+ or 21.0.6+ on all Java application servers. Deploy WAF virtual patching for Base64 input validation as an interim control if immediate patching is not possible.
- Audit third-party JavaScript on all web properties. Implement Subresource Integrity hashes and strict Content Security Policy directives. Monitor CSP violation reports for anomalous script sources.
- Expand IR playbooks to include downstream partner notification when a cloud tenant breach exposes shared data. Revoke OAuth tokens and shared document links associated with compromised accounts.
- Patch Puppet Server for CVE-2026-115731. Audit and revoke unnecessary client certificates. Segment Puppet server network access to authorized management subnets only.
- Check PyPI dependencies for Microsoft agent-framework packages - versions 1.26.1 and 1.26.2 are poisoned with credential stealers targeting Azure and GitHub Actions secrets. Pin to known-good versions and audit CI/CD pipeline secrets for exposure.
Detection Spotlight
Java4Shell exploitation will appear as malformed Base64 input in HTTP request bodies or headers hitting Java application endpoints. The following Splunk SPL query identifies unusually long or malformed Base64 strings in web server access logs that may indicate exploitation attempts against CVE-2026-114938. Tune the base64_length threshold based on your application’s normal Base64 payload sizes - most legitimate Base64 tokens are under 2048 characters.
index=web sourcetype=access_combined
| rex field=_raw "(?i)(Authorization|Content-Body|X-Token):\s*(?:Basic\s+)?(?<b64_value>[A-Za-z0-9+/=]{64,})"
| eval b64_len=len(b64_value)
| where b64_len > 2048 OR match(b64_value, "[^A-Za-z0-9+/=]")
| stats count by src_ip, uri_path, b64_len
| where count > 5
| sort -count
This query extracts Base64 values from common headers and request fields, flags those exceeding 2048 characters or containing invalid Base64 characters (which would trigger the vulnerable SIMD code path), and groups by source IP and target URI. A high count from a single source against authentication or API endpoints warrants immediate investigation. False positive rate is low in environments where Base64 payloads are well-understood; expect tuning in environments with large file upload endpoints.
References
- CISA Java4Shell Advisory - The Record
- Java4Shell Technical Analysis - Cisco Talos
- Optimizely Supply-Chain Attack Investigation - Krebs on Security
- Cold Relic Google Workspace Chain - The Hacker News / Mandiant
- CVE-2026-115731 Puppet Server RCE - SecurityWeek
- UAT-12643 Overlord RAT Blockchain C2 - Cisco Talos
- Microsoft Agent Framework PyPI Poisoning - SecurityWeek
- Uber Driver Data Breach - BleepingComputer
Related Briefs
- Optimizely Supply-Chain Attack Hits 178 Orgs
- AI Agents Run Recursive Key-Theft Loops
- FBI Warns ARTEX AI Powers MSP Supply-Chain Hits
- AntiFox AI Agent Hits AWS Secrets in 30 Minutes
- Check Point Zero-Day Chains Root Access With ESXi Wipes
Subscribe to it-learn Brief
Get it-learn Brief in your inbox (Mon–Fri) - Daily cybersecurity news, SE angles, and detection queries.