Skip to content

Response, New York. Noah Park: Check Point Patches Actively Exploited Management Server Path Traversal. Detection, Dallas. Ava Okello: Report Finds Nearly Half of Deployed Detections Need Attention Before SOCs Can Trust Them.

SOCtember

Always First. Fast SOC News.

Cloudflare Radar Zero Wire briefing card titled Java4Shell: remote code execution in OpenJDK, CVE-2026-114938, with a satellite dish illustration.
Image: Cloudflare (fair use). The Cloudflare Radar Zero Wire briefing card for CVE-2026-114938.

Detection

The First Java4Shell Attack Attempts Are Here: GreyNoise Logs 187 Booby-Trapped Login Headers, Cloudflare Sees Scanners Return Disguised as Chrome, and SOC Tools Are on the Affected Lists

Two days after the OpenJDK Base64 flaw went public, sensors are catching correctly shaped trigger strings aimed at Confluence and Spring Boot login paths. No break-in has been confirmed, but the fingerprint is narrow enough to hunt today, and Elasticsearch, Logstash and Cisco Secure Network Analytics are among the products that need the fix.

Ava Okello, DetectionLondon7 min read

LONDON - The first attempts to exploit Java4Shell, the heap overflow in the Java runtime's Base64 decoder that was disclosed on Tuesday, have started arriving at internet sensors, according to threat intelligence firms tracking the flaw, which said on Thursday that none of the attempts they had seen had yet produced a confirmed break-in.

GreyNoise, which runs a network of decoy servers, said on a public tracking page last updated at 10:48 UTC on Thursday that it had recorded 187 requests carrying the Java4Shell trigger shape from five IP addresses since the start of Tuesday, with zero successful intrusions against its sensors. Cloudflare's threat research team, Cloudforce One, said scanners were already hiding what they were.

The flaw, tracked as CVE-2026-114938, was found by the security firm Calif and patched by Oracle and the OpenJDK project on Nov. 17. Defenders spent the first two days on inventory. The arrival of shaped attack traffic gives security operations teams something more concrete: a request fingerprint narrow enough to search for in existing web and proxy logs.

What the attack traffic looks like

The bug sits in hand-written machine code that the HotSpot Java virtual machine uses to decode Base64 quickly on x86-64 processors with AVX-512 VBMI instructions, which means Intel Ice Lake or newer and AMD Zen 4 or newer. Cloudflare explained that when padding characters appear where valid Base64 never places them, the routine computes a negative output length and writes a full 64-byte block anyway. For a 65-character input made of 64 characters followed by a single equals sign, OpenJDK allocates a 50-byte array and writes 62 bytes past its end. The simplest trigger, the company said, is 64 letter A's followed by "=". Java rejects the input as invalid, but only after the neighboring memory has been overwritten, so an application's own validation does not help.

GreyNoise defines an attempt as an HTTP Authorization header whose Basic or Bearer value has that shape: 64 or 128 characters plus one to three more, ending in "=". The busiest source on its page, an address on Swisscom's network in Switzerland that GreyNoise first tracked on Thursday morning, sent 70 sessions requesting /rest/api/space and /login.action, two Confluence paths, along with /actuator/env, a Spring Boot endpoint. Another source requested /rest/api-application/1.0/application-info and /rpc/xml. GreyNoise tags the traffic "Base64 Heap Overflow RCE Attempt v2" and lists 12 addresses as block candidates.

Screenshot of the GreyNoise Java4Shell activity page showing 5 IPs sending Java4Shell-shaped Base64, 187 Java4Shell-shaped requests and 0 successful intrusions, with a timeline of the disclosure and patch release.
Image: GreyNoise (fair use).The GreyNoise Java4Shell tracking page as of Thursday morning, Nov. 19.

Cloudforce One reported two shifts on Wednesday. At 06:59:34 that day, it said, it saw "the first instance of ARTEX and Java4Shell being used from the same IP address," a reference to the autonomous attack framework that U.S. agencies warned about last week. Late Wednesday it added that "scanners are starting to hide their tool names": one address first probed a target with an obvious java4shell user agent, then about 17 hours later "sent the exploit trigger to the same target, this time posing as Chrome."

Screenshot of a Cloudforce One post on X from Nov. 18, 2026, saying scanners are starting to hide their tool names and that one IP returned about 17 hours after a java4shell user-agent probe, posing as Chrome.
Image: Cloudforce One on X (fair use).Cloudflare's threat research team on the first disguised Java4Shell traffic, Nov. 18.

That second observation matters for detection. A rule keyed to the java4shell user agent string will see the reconnaissance and miss the attempt that follows. The durable signal is the shape of the Base64 value itself, not who claims to be sending it.

The watchTowr honeypot network is seeing the same pattern. Jake Knott, its head of threat intelligence, said in comments published by SecurityWeek that attackers were spraying requests "decorated in 'java4shell-recon' User-Agents, alongside blind exploitation attempts using correctly shaped requests," and that he expected them "to iterate product-by-product" toward more reliable exploits. Previdian, another sensor operator, said it had logged nine exploitation attempts from six addresses on Wednesday and Thursday but could not confirm any compromise, according to SecurityWeek. On Wednesday evening, Johannes Ullrich of the SANS Internet Storm Center summed up his own Java honeypots this way: "We are seeing many attacks, but none that exploit Java4Shell (yet)."

Why the SOC's own tools are in scope

The affected lists now include software that security teams run themselves. Elastic said Elasticsearch 7.17.3 and later is vulnerable on affected CPUs, with remote code execution possible for an authenticated user holding at least read access and denial of service possible without authentication. Logstash 8.6.2 and later is exposed only through the Elastic Integration filter when an ingest pipeline decodes Base64 from untrusted events. Kibana, Beats, Elastic Agent and APM Server do not use Java and are not affected. Fixed builds are due in Elastic 8.19.26, 9.4.10, 9.5.7 and 9.6.0.

Cisco's advisory, first published Wednesday, lists Secure Network Analytics, Secure Workload, Catalyst Center and Splunk AppDynamics On-Premises Virtual Appliance among the affected products, with fixes still to come and no workarounds. It is still investigating IOS XE and UCS Central.

Symantec noted that Base64 is everywhere a login or API service touches untrusted data, including authentication headers, JSON Web Tokens and SAML assertions, and that a failed attempt can crash the JVM, although Calif reported that retries often succeed. Calif has also described a client-side variant, in which a Java program that connects over TLS to a malicious server can be compromised.

What to hunt for, and what a WAF cannot see

For teams with header logging at a reverse proxy, load balancer or WAF, the GreyNoise definition translates into a simple search: Authorization values of 65 to 67 or 129 to 131 characters ending in "=", especially against Confluence, Jira, Jenkins, Keycloak, Spring Boot actuator and API paths. Unexpected restarts or crash logs from Java services on AVX-512 hardware since Tuesday deserve a second look, given Symantec's warning that failed attempts can take the JVM down.

Cloudflare has shipped six managed WAF rules covering malformed Base64 in headers, query values, request bodies, path segments, JWT components and WebAuthn values. They are enabled in Log mode by default because some benign token formats look similar, and the company recommends reviewing matches before switching them to Block. It was blunt about the limits: a WAF will not see Base64 that is wrapped in another encoding, traffic over non-HTTP protocols such as the Kafka broker's, internal service-to-service calls, or the outbound TLS path in the client-side variant.

Screenshot of the Cloudflare blog section on new managed rules for Java4Shell, listing rule descriptions and rule IDs for malformed Base64 in headers, query values, body values and path segments.
Image: Cloudflare (fair use).Cloudflare's managed WAF rules for CVE-2026-114938, from its Nov. 18 blog post.

The fix is the patched runtime, followed by a restart, because upgrading a package does not replace a JVM that is already running and many products bundle their own Java. Where patching has to wait, Calif documented two startup flags that remove the vulnerable code path, -XX:-UseBASE64Intrinsics and, on x86-64 only, -XX:UseAVX=2, both at some cost in performance. Cloudflare said it began rolling the first of those out across its own Java workloads about an hour after the patches were published and has seen no sign of exploitation on its infrastructure.

About the Author

Ava Okello

Ava Okello covers detection engineering, EDR telemetry, and SOC hunting for SOCtember from London.

Related stories

Detection desk