Security news, decoded.

The Dutch Institute for Vulnerability Disclosure (DIVD) says it suffered a breach on September 21 after attackers used an agentic AI system to exploit two previously unknown flaws in the open-source helpdesk platform Zammad. DIVD describes the intrusion as unlike anything it has handled before, noting the automated nature of the attack chain. The first bug, CVE-2026-102489 (CVSS 9.4), lets unauthenticated attackers hijack sessions and execute code remotely, while the second, CVE-2026-102490 (CVSS 9.4), allows a local account to escalate privileges to root; chained together, these flaws let the intruders move from ordinary Zammad access to full root control within seconds. The attackers reportedly pivoted toward other connected services and pulled out some data, though network segmentation limited further damage, and DIVD says it is still treating the incident as a confirmed breach while investigating remaining signs of compromise. Zammad versions 6.3.0 through 6.5.4 are vulnerable, and while 7.0.0 through 7.1.3 technically contain the same bug, real-world conditions block exploitation there; DIVD urges all Zammad users to update to version 7 or take affected instances offline, and has released a detection script plus is actively scanning the internet to warn exposed organizations.

CISA has placed a severe authentication-bypass vulnerability in Cisco Catalyst SD-WAN Manager on its Known Exploited Vulnerabilities list after confirming it is being actively abused in the wild. The flaw, CVE-2026-76504, carries a CVSS score of 9.8 and stems from faulty URI hex-encoding handling in HTTP requests, letting an unauthenticated remote attacker send a crafted request and reach the system's API with full admin rights. Cisco disclosed in September 2026 that it had detected exploitation but has not shared details on the attackers, the scale of compromise, or when the activity began, though it did publish log-based indicators of compromise tied to suspicious jsecuritycheck entries. Federal civilian agencies must apply fixes by October 3, 2026, while watchTowr notes this marks the eighth Cisco SD-WAN CVE added to KEV in 2026 alone, underscoring why organizations should urgently patch and audit logs for exploitation attempts.

Cisco has confirmed that a new flaw in its SD-WAN platform, tracked as CVE-2026-76504, is being actively abused by attackers — the fifth such zero-day disclosed by the company this year. The company's security team spotted the active exploitation in September 2026 while investigating a customer support ticket, though it hasn't yet disclosed details about the attackers or their methods beyond sharing indicators of compromise. The bug sits in Cisco Catalyst SD-WAN Manager, the console used to administer SD-WAN deployments, and stems from how the system processes URI encoding in HTTP requests. This flaw lets an attacker send a specially crafted request that slips past an authentication check meant to guard a particular API endpoint, effectively granting them admin-level API access and, potentially, control over the entire network. It affects Manager versions 26.2, 26.1, 20.18, 20.15, 20.12, 20.9 and all earlier releases, no matter how the system is configured. Patches are available for the 20.9 through 26.2 branches, while customers on anything older than 20.9.10.1 need to upgrade since no workaround exists. Cisco recommends shielding on-premises installations from the public internet and, if remote access is unavoidable, limiting it to known trusted hosts. Organizations are also urged to check their serviceproxy-access.log and vmanage-server.log files for the published IoCs — and to save logs before patching in case forensic investigation is needed later. CISA has added the vulnerability to its Known Exploited Vulnerabilities list, giving US federal agencies until October 3, 2026, to patch and conduct a compromise assessment.

Cisco has issued emergency fixes for a critical flaw in Catalyst SD-WAN Manager that attackers are already exploiting. The bug, CVE-2026-76504, scores 9.8 on CVSS and stems from flawed URI encoding handling in HTTP requests, letting unauthenticated remote attackers slip past API session authentication and reach a restricted endpoint with admin-level privileges. Every Catalyst SD-WAN Manager deployment is vulnerable regardless of setup, and no workaround exists, so Cisco urges immediate upgrades to versions 26.2.1, 26.1.2.1, 20.18.4.1, 20.15.6.1, 20.12.8.2, or 20.9.10.1 — fixes already applied to Cisco-hosted SD-WAN instances. CISA has added the flaw to its Known Exploited Vulnerabilities catalog with a three-day remediation deadline for federal agencies, though neither Cisco nor CISA has disclosed specifics about the ongoing attacks.

Researchers at Calif have released a working proof-of-concept for CVE-2026-86950, a CoreGraphics memory-corruption bug that Apple says was likely exploited in a highly sophisticated attack against select targeted individuals on iOS versions prior to iOS 27. The bug stems from inconsistent handling of out-of-range glyph coordinates across CoreGraphics' rasterizer functions, letting a specially crafted font embedded in a PDF trigger an out-of-bounds write and crash both iPhones and Macs; the researchers only demonstrate the crash, not full code execution. Apple fixed the issue on September 28, crediting Meta Product Security, and CISA added the flaw to its Known Exploited Vulnerabilities catalog the next day, giving federal agencies until October 2 to patch. Because Meta was credited with the find, Calif also looked at WhatsApp's attachment scanner and found new code flagging suspicious font streams in PDFs, which it called circumstantial evidence of a possible delivery route — though an initial, more specific claim about a WhatsApp zero-click chain was quietly removed from the published writeup shortly after release, and neither WhatsApp nor Meta has confirmed any connection.

Google Threat Intelligence Group studied vulnerability disclosures and exploitation from January 2025 through August 2026 and found that flaws discovered by AI research agents are weaponized unusually fast. A standout case is CVE-2026-1731, an unauthenticated OS command injection bug in BeyondTrust Privileged Remote Access and Remote Support identified by the Hacktron AI agent: one threat cluster began exploiting it within four days of public disclosure, and five more clusters followed within a week, deploying SNOWLIGHT, SPARKRAT, and cryptominers after escalating privileges and stealing data. GTIG also reports that half of the vulnerabilities it attributes to AI discovery result in remote code execution, versus just 26% of those found through other methods, and that overall exploited CVE counts in 2026 already surpass all of 2025. AI infrastructure itself is increasingly targeted: agent orchestration tools like Flowise and Langflow account for half of this year's AI-related disclosures, while inference platforms including vLLM, Ollama, and LiteLLM saw actively exploited bugs such as CVE-2026-42271, CVE-2026-5027, and CVE-2025-3248, enabling host takeover, credential theft, and remote code execution.

LevelBlue's threat hunting team has uncovered active abuse of a critical pre-authentication command injection flaw in Citrix NetScaler ADC and NetScaler Gateway, tracked as CVE-2026-88771 (CVSS 9.5), which lets unauthenticated attackers run arbitrary commands. Analysts found malicious authentication attempts across several customer environments containing strings tied to the exploit, along with follow-on activity fetching second-stage payloads via curl or wget from external servers. One Perl payload adds a hidden superuser account named secmonitor to the device configuration, archives and exfiltrates NetScaler config data, then plants a PHP web shell disguised with URLs mimicking legitimate NetScaler CSS files before erasing its own traces. A separate Python payload opens a reverse shell over port 443 and kills a monitoring process to stay hidden. The findings follow recent disclosures that a related flaw, CVE-2026-88772, has already been exploited against dozens of organizations to deliver web shells and tunneling tools, underscoring the urgency for Citrix NetScaler customers to patch and hunt for compromise indicators.

Microsoft has disclosed that attackers have been actively exploiting a severe flaw in Synacor's Zimbra Collaboration Suite, tracked as CVE-2026-73570, to steal email archives and login credentials from vulnerable mail servers. The bug allows unauthenticated remote attackers to execute operating system commands by sending a specially crafted email that abuses the SNMP notification feature, provided the optional zimbra-snmp package is installed and enabled. Although Synacor released a fix on July 20, the vulnerability wasn't publicly disclosed until roughly three weeks later, giving attackers a window to scan for and compromise targets. The Shadowserver Foundation reports 274 compromised instances out of around 10,000 still-exposed Zimbra deployments, with Microsoft observing web shells, reverse shells, privilege escalation, and mass exfiltration of mailbox data across multiple industries and regions.

The Dutch vulnerability disclosure nonprofit DIVD has revealed that attackers compromised its own network by chaining two previously unknown flaws in the open-source Zammad helpdesk software, tracked as CVE-2026-102489 and CVE-2026-102490. Together these bugs allowed session takeover, remote code execution, and privilege escalation from a regular Zammad account to full root access, all carried out within seconds by an autonomous AI agent rather than a human operator. Because the agent logged its own reasoning during the intrusion, DIVD was able to reconstruct exactly how the breach unfolded, though network segmentation and quick response reportedly stopped the attacker from reaching deeper into the organization's systems. DIVD, working with Merlon Security, has notified Zammad and is urging the platform's roughly 2,000 customers and 55,000 users to upgrade to Zammad version 7 or take vulnerable instances offline immediately.

Microsoft's security research team reports that attackers have actively abused a now-fixed vulnerability in Synacor's Zimbra Collaboration Suite to plant web shells and steal mailbox and authentication data. The flaw, CVE-2026-73570 (CVSS 8.9), is an unauthenticated OS command-injection bug triggered via a crafted SMTP email when SNMP notifications and the optional zimbra-snmp package are active, requiring no login or user interaction. Zimbra shipped a fix in version 10.1.20 in July 2026, but telemetry shows exploitation occurred between that release and the public disclosure in mid-August, with intruders deploying JSP web shells, reverse shells, cron/systemd persistence, custom Go-based implants (Zimdown2/Zimclient2), and tools to dump LDAP secrets, mailbox databases, and attempt cloud-based exfiltration via AzCopy. CISA has since added the bug to its KEV catalog, and defenders are urged to patch immediately, remove the SNMP package if unpatched, restrict SMTP/SNMP exposure, and rotate Zimbra credentials.

CISA has issued an alert about a severe flaw in MikroTik RouterOS, tracked as CVE-2026-84411, that lets an attacker send a single malicious request to the device's web-management interface without needing to log in. The bug is an integer underflow in how RouterOS parses HTTP request bodies, and it can be abused to run arbitrary code with root privileges or to knock the device offline. CISA says versions prior to 7.24 are impacted, though the vendor's guidance points to updating to 7.23 or newer; the current stable release is 7.24.4 and the latest long-term build is 7.23.7. There's no confirmed exploitation yet, but MikroTik devices are a frequent target for botnets and attackers, including a recent campaign chaining separate RouterOS flaws (CVE-2026-67276 and CVE-2026-86060) to seize routers exposing SSH to the internet.

Cisco has disclosed that a newly found critical flaw in Catalyst SD-WAN Manager, tracked as CVE-2026-76504, is already being abused by attackers in real-world attacks. The bug, scoring 9.8 on CVSS, stems from improper handling of URI encoding in login-session requests, letting an unauthenticated attacker craft a request that slips past an access restriction on a single API endpoint and operate the Manager's API as the admin account, which carries the powerful netadmin role by default. Cisco's incident response team learned of the active abuse in September 2026 while reviewing a TAC support case, though it has not disclosed how many organizations were hit or who is behind the attacks. There is no workaround; fixed builds exist for every release train (for example 20.9.10.1, 20.12.8.2, 20.15.6.1, 20.18.4.1, 26.1.2.1 and 26.2.1), and customers are urged to upgrade quickly, restrict internet exposure to the Manager, and check logs for suspicious jsecuritycheck entries that may indicate compromise.

A new Google Threat Intelligence Group (GTIG) study shows that AI is reshaping both the speed and nature of vulnerability research. Comparing data from January 2025 to August 2026, GTIG found monthly disclosures more than doubled, climbing from 5,045 in January 2026 to over 10,700 by August, while in-the-wild exploitation nearly doubled too, reaching an average of 18 exploited flaws per month versus 10.5 the previous year. GTIG notes that much of the raw disclosure growth stems from automated CVE assignment in open-source projects — Linux Kernel-related entries alone account for roughly 5,000 CVEs with no observed zero-day exploitation. More significant, the group says, is a 167% rise in high-risk disclosures and a surge in exploitation of n-day flaws, suggesting attackers are increasingly using AI to speed up patch-diffing and weaponize known bugs rather than hunt fresh zero-days. The report also highlights that AI-discovered vulnerabilities carry a different risk profile, more often enabling remote code execution, and cites real-world exploitation of AI-found bugs such as a BeyondTrust flaw exploited within days of disclosure. Separately, GTIG tracked over 2,000 CVEs tied to AI systems themselves since 2025, with confirmed in-the-wild attacks hitting tools like LiteLLM and Langflow, underscoring growing risk across AI infrastructure.

WatchGuard has released updates addressing 15 vulnerabilities in Fireware OS, the most serious being a critical code-injection flaw tracked as CVE-2026-86131 with a CVSS score of 9.2. The bug lies in how the OS processes BOVPN over TLS client settings, letting an attacker controlling the remote VPN server run commands as root on a connected Firebox device. The fix ships in Fireware OS 2026.3.2, 2026.2.3, 12.12.3, and 12.5.21, alongside patches for 13 high-severity issues (covering RCE, auth bypass, DoS, unauthorized SSLVPN access, and arbitrary file reads) and one medium-severity authorization flaw affecting web app access. A day earlier, WatchGuard also patched two critical and one high-severity flaw in its Access Point firmware — CVE-2026-101891 and CVE-2026-86102 allow unauthenticated attackers to hijack API sessions and run shell commands, while the high-severity bug needs admin rights for OS command injection; all three are fixed in WatchGuard AP 3.4.8. The vendor says none of these flaws have been exploited in the wild so far.

Google's Mandiant and GTIG have detailed an espionage campaign that abused two Citrix NetScaler zero-days, CVE-2026-88771 and CVE-2026-88772, for weeks before patches became available over the weekend. Both unauthenticated remote-code-execution flaws hit NetScaler ADC and NetScaler Gateway appliances, and investigators say exploitation began as early as September, well before public disclosure. Suspected state-sponsored actors used the bugs to gain root access, altered the web server configuration to plant root-privileged web shells, and deployed previously unseen tools — a PHP shell called WHIPSHOT and a Python tunneling utility named SLAPSHOT — to pivot into victims' internal networks for reconnaissance, lateral movement and credential theft. Targets span government, financial, education, legal and professional-services organizations across North America and Europe, with researchers tracking dozens to over 100 compromised networks and tens of thousands of exposed instances, prompting urgent calls for immediate patching.

Mandiant and Google Threat Intelligence Group report that suspected state-sponsored hackers have been actively abusing CVE-2026-88772, one of two newly disclosed Citrix NetScaler ADC and NetScaler Gateway zero-days, since at least early September 2026. The flaw, which only affects appliances with DTLS configuration enabled, lets attackers bypass authentication by sending malformed or fragmented record headers that corrupt heap memory in the packet processing engine, granting root-level access on the underlying FreeBSD system. Dozens of organizations across North America and Europe — spanning government, finance, education, telecom, and legal sectors — have reportedly already been compromised, with intruders deploying web shells and a custom tunneling tool called SLAPSHOT to pivot into internal networks. Mandiant's CTO warns that simply patching NetScaler devices against CVE-2026-88772 and the companion flaw CVE-2026-88771 won't remove attackers who already gained footholds or address exposed credentials, urging full incident-response and remediation steps instead.

TeamViewer has pushed out an unusually urgent advisory telling customers to update their Full Client and Host software without delay after discovering five high-severity vulnerabilities. The most dangerous, CVE-2026-92370, is an access-control weakness in the Windows, Linux, and macOS clients that could allow a remote attacker to bypass session restrictions and ultimately execute code on a target machine. The remaining four bugs — a path traversal flaw (CVE-2026-19743), a heap buffer overflow (CVE-2026-92368), a race condition (CVE-2026-92369), and a path validation issue (CVE-2026-92371) — could let a local attacker run code with the logged-in user's rights or escalate to SYSTEM or root privileges. TeamViewer says it has seen no public exploit code or in-the-wild attacks so far, but is urging everyone to move to version 15.82 or the appropriate maintenance release, especially given the software's history as a target for ransomware crews and, in a prior breach, the Russian state-linked group Midnight Blizzard.

Google and Mozilla both pushed out major browser updates this week, collectively closing more than 100 security holes. Chrome's new build, versions 154.0.8037.92/.93 for Windows and macOS and 154.0.8037.92 for Linux, fixes 32 issues, among them a critical buffer overflow in ANGLE identified as CVE-2026-102331, along with 25 high-severity bugs such as use-after-free and uninitialized resource flaws and five type confusion problems in the V8 engine. Mozilla's Firefox 157 tackles roughly 76 vulnerabilities, including 38 high-severity ones like use-after-free and sandbox escape bugs, with many of the same fixes also backported to Firefox ESR 153.4, 140.17, and 115.42. Neither company reports that any of these flaws have been exploited in real-world attacks, but both urge users to update their browsers promptly given the scale and severity of the fixes.

OpenInfra Europe, the European arm of the OpenInfra Foundation that backs projects like OpenStack, has disclosed that its self-hosted JFrog Artifactory repository at artifactory.nordix.org was breached. The group is urging anyone who pulled artifacts from that server between August 28 and September 15, 2026 to treat them as potentially tampered with, pull them from build pipelines immediately, and stop relying on them. According to the notice, attackers exploited CVE-2026-82329, an authentication-bypass flaw in JFrog Artifactory that let unauthenticated actors obtain administrator rights and manipulate stored packages, credentials and integrations. JFrog disclosed the bug on August 28, 2026, CISA added it to its Known Exploited Vulnerabilities list on September 2, and outside researchers report active exploitation starting August 31 — matching when OpenInfra Europe says its own instance was first compromised. The organization says it only noticed the intrusion on September 15, after a legitimate user was locked out, prompting it to isolate the server and launch an investigation whose full scope remains unclear. Any team consuming artifacts from this repository during the exposure window should audit their pipelines and rotate affected credentials as a precaution.

Mandiant and Google's Threat Intelligence Group report that unidentified attackers are actively exploiting a recently fixed vulnerability in Citrix NetScaler ADC and NetScaler Gateway appliances, hitting government, financial, tech, education and legal organizations across North America and Europe since September 2026. The flaw, CVE-2026-88772 (CVSS 9.5), is a memory overflow in how the NSPPE component handles DTLS protocol records, letting attackers skip authentication and crash the packet engine to seize root-level control of the FreeBSD-based appliance. Successful compromise leads to deployment of a new PHP web shell called WHIPSHOT, which hides Base64-encoded commands inside HTTP headers, alongside a Python tunneling tool named SLAPSHOT that proxies traffic into internal networks for reconnaissance and credential theft, with both tools disguised as .deb or .sig files and configured to self-delete after inactivity. GreyNoise separately reported a sharp jump in mass exploitation of CVE-2026-88771 and CVE-2026-88772 by multiple independent actors starting September 28, 2026, warning that administrators of internet-facing NetScaler gear should patch immediately given the scale of ongoing attacks.

OpenSSL has released updates addressing a High-severity vulnerability, CVE-2026-84782, found in its DTLS implementation. The bug can occur when a retransmission timer fires while a large handshake message is still being sent in pieces, causing the resend to grab the wrong buffer position—this can expose unencrypted heap data to the remote peer or crash the application by reading into unmapped memory. There is no indication the flaw has been exploited, and no workaround exists for systems that cannot immediately upgrade. The fix is included in OpenSSL versions 4.0.3, 3.6.5, 3.5.9, and 3.4.8, while patches for the older 3.0, 1.1.1, and 1.0.2 branches are available only through paid premium support, since 3.0 lost public security updates on September 7. CISA scored the issue 8.2 out of 10, and Linux distributions including Ubuntu and Debian have already shipped their own updated packages. The same release batch also resolves 13 lower-severity issues, among them a Moderate flaw (CVE-2026-84783) that could crash multi-threaded TLS clients or servers under specific certificate-chain conditions.

OpenSSL and WolfSSL have each shipped fixes for roughly a dozen vulnerabilities, with several rated high severity. The most serious OpenSSL bug, CVE-2026-84782 (CVSS 8.2), affects applications using DTLS—common in VPN, VoIP and IoT software—and lets an unauthenticated remote peer leak fragments of heap memory in plaintext or crash the application during handshake retransmission. A second, medium-severity OpenSSL flaw, CVE-2026-84783, can crash multi-threaded TLS clients, while the remaining lower-severity issues cover DoS via resource exhaustion, QUIC-based DDoS amplification, and timing side-channels that could expose private keys. Separately, WolfSSL 5.9.4, released September 25, patches 11 vulnerabilities including three high-severity authentication-bypass issues—CVE-2026-93302, CVE-2026-89102 and CVE-2026-89136—that could let malicious servers or attackers holding a chained certificate forge identities or impersonate trusted peers in certain configurations. No exploitation in the wild has been reported for either library, but given their widespread use, administrators are advised to update promptly.

Security researchers at watchTowr have published a detailed technical breakdown of CVE-2026-88772, a critical flaw affecting Citrix NetScaler ADC and NetScaler Gateway that attackers are already exploiting in the wild. The bug stems from how the NetScaler Packet Processing Engine handles DTLS handshake reassembly: it trusts a tiny declared fragment size while the overall message length field claims a much larger total, letting an attacker craft fragmented records that overflow a fixed 35,840-byte scratch buffer with up to roughly 174 KB of data. According to researcher Sina Kheirkhah, this overflow can be chained with an mprotect() call to bypass NX protections and redirect execution to attacker-controlled shellcode with root privileges, all without authentication. The disclosure follows closely on the heels of a related PoC for CVE-2026-88771, which has reportedly been used together with CVE-2026-88772 in active attack campaigns, underscoring the urgency for Citrix customers to apply available patches immediately.

Apple has pushed out iOS and macOS updates to fix an out-of-bounds write bug in its CoreGraphics framework, tracked as CVE-2026-86950, which a threat actor has already used in real-world attacks. The flaw carries a CVSS score of 8.8 and lets an attacker run arbitrary code, affecting a wide swath of devices from the iPhone 11 onward as well as multiple iPad generations, with Apple also issuing fixes for macOS Tahoe and Sequoia. Apple describes the activity as an "extremely sophisticated" campaign aimed at a narrow set of specific individuals, language researchers say suggests a nation-state or commercial spyware operator rather than ordinary cybercriminals. CISA has added the flaw to its Known Exploited Vulnerabilities catalog and, under Binding Operational Directive 26-04, is requiring federal agencies to apply Apple's mitigation within three days and complete a compromise check by October 2, while experts urge all organizations to treat the patch as an emergency update rather than wait for routine cycles.

Security researcher Ariel Fogel of Pillar Security disclosed a vulnerability in Unsloth Studio, the web interface for the popular Unsloth library used to fine-tune and quantize LLMs, that let a malicious model repository run arbitrary Python code merely by having its configuration inspected. The root cause was Unsloth Studio's use of the trustremotecode=True setting in Hugging Face Transformers, which let custom code referenced in a model's config.json execute before the model weights were ever loaded or any inference was run. An attacker exploiting this could run code with the victim's privileges, potentially stealing training data, model artifacts, cloud credentials, or SSH keys from development environments. Pillar reported the issue in early June and Unsloth shipped a fix in version 2026.6.9, though the two sides disagreed on severity and no CVE was ultimately assigned; Pillar says it has seen no in-the-wild exploitation but notes similar trustremotecode issues have affected other AI tools this year, including LMDeploy (CVE-2026-46432), vLLM (CVE-2026-4944), and InstructLab (CVE-2026-6859).

Security researchers report active exploitation of two Citrix NetScaler zero-day vulnerabilities, CVE-2026-88771 and CVE-2026-88772, affecting NetScaler ADC and Gateway appliances. According to Mandiant, attacks began as early as September and have hit government, financial, education, legal, and professional-services organizations across North America and Europe, with attackers gaining root access, planting web shells, and moving laterally inside victim networks. GreyNoise independently spotted exploitation of a NetScaler Gateway days before Citrix's public disclosure, observing attempts to drop a hidden PHP web shell and tamper with shell permissions to maintain elevated access. Mandiant also identified two new malware families, WHIPSHOT and SLAPSHOT, used together as a proxy-and-tunneling toolkit to steal credentials and reach internal systems, and urges admins to apply Citrix's latest patches immediately since mitigations like disabling DTLS only cover one of the two flaws.

Researchers from VUsec and Scuola Superiore Sant'Anna disclosed a new speculative-execution attack dubbed Branch Target Reuse (BTR), a variant of Spectre v2, tracked as CVE-2026-64507 and CVE-2026-64508. The technique abuses stale entries in a CPU's branch predictor left over when a JIT engine reuses memory for new self-modifying code, tricking the processor into briefly executing attacker-controlled instructions and leaking data through cache side channels. Using unprivileged classic BPF programs on Linux, the team showed they could pull a running 'su' process's root password hash out of memory at roughly eight bytes per second, fully recovering it in about 3 to 5 minutes on Intel's Raptor Cove and Lion Cove cores, and they confirmed the same weakness affects Intel, AMD and Arm processors. Firefox's SpiderMonkey and Oracle's GraalVM were also found susceptible to related stale-prediction effects, though no full exploit chain was completed there. Fixes have already landed in the Linux kernel, and users are urged to install the latest OS, firmware and kernel updates.

Researchers from VUSec and Scuola Superiore Sant'Anna have published a new Spectre v2 branch-prediction attack they call Branch Target Reuse (BTR), which works against JIT compilers even when existing speculative-execution protections are in place. The flaw stems from CPUs failing to clear stale indirect-branch-prediction entries when self-modifying JIT code is freed and reallocated, letting an attacker redirect transient execution to invalid code locations and leak secrets. The team demonstrated working exploits against the Linux kernel's cBPF JIT, Mozilla Firefox's SpiderMonkey engine, and Oracle's GraalVM, including two proof-of-concept attacks that recovered a root password hash within minutes on a fully updated, default-hardened Intel Linux machine. The issue has been assigned CVE-2026-64507 and CVE-2026-64508, with kernel fixes already merged, while GraalVM now randomizes JIT code-cache placement and Mozilla is leaning on site isolation rather than IBPB-based mitigation.

Attacks against Citrix NetScaler ADC and Gateway appliances have surged from quiet, targeted zero-day abuse into indiscriminate mass scanning, after a technical write-up and a working exploit for CVE-2026-88771 were made public. Citrix had already confirmed active exploitation when it released patches for eight flaws, including CVE-2026-88771 and CVE-2026-88772, both abused before disclosure, and issued a (admittedly imperfect) compromise-detection script. Researchers from GreyNoise, Lupovis, CERT-EU and Censys report attackers probing for admin access, planting webshells disguised as stylesheets, erasing logs, and exfiltrating data to a Hetzner-hosted server, with roughly 42,000 NetScaler instances still reachable online and fewer than 10% of them patched. One researcher says over 100 victim organizations have already been identified, each with a custom backdoor, pointing to likely espionage motives behind the initial wave of attacks. No public exploit exists yet for CVE-2026-88772, but defenders are urged to patch immediately and hunt for the indicators shared by the security community.

Citrix confirmed on September 27 that two critical NetScaler zero-day vulnerabilities, CVE-2026-88771 and CVE-2026-88772, were being actively exploited on unmitigated appliances, finally addressing rumors that had circulated for days. CVE-2026-88771 is an input-validation flaw enabling remote code execution, while CVE-2026-88772 is a memory overflow bug that can also trigger DDoS attacks; both carry a 9.5 CVSS score and hit default configurations of NetScaler ADC and Gateway. Researchers, including watchTowr's Benjamin Harris and GreyNoise, say attacks were underway at least a week before Citrix's advisory, with patch metadata suggesting Citrix knew earlier, and Palo Alto Networks scans found over 50,000 exposed NetScaler instances online. Because exploitation is trivial against default setups, experts recommend organizations unable to immediately patch should consider taking affected NetScaler devices offline.

watchTowr Labs has detailed a second zero-day from Citrix's latest NetScaler bulletin, CVE-2026-88772, which sits alongside the command-injection flaw CVE-2026-88771 disclosed in the same CTX697096 advisory covering eight bugs in total. This issue is a preauthentication memory overflow in the DTLS handling of NetScaler ADC and NetScaler Gateway, triggered when an attacker sends crafted DTLS handshake fragments that abuse inconsistent length fields to smuggle roughly 174KB of data into a fixed 35,840-byte scratch buffer inside the nsppe process. Because DTLS is enabled by default on VPN virtual servers, no authentication is required, and Citrix confirms the bug (CVSS 9.5) has already been exploited as a zero-day, researchers managed to turn the overflow into full remote code execution via a ROP chain and mprotect call. Citrix, now under Cloud Software Group, urges customers to upgrade to NetScaler 14.1-73.37, 13.1-64.23, or the corresponding FIPS/NDcPP builds without delay.

Dutch authorities announced the arrest of a 24-year-old man from Amsterdam suspected of ties to the ShinyHunters extortion crew. The suspect, reportedly Pepijn van der Stap, had already been convicted in 2023 for hacking intrusions, data theft and extortion, and was arrested again on September 16 while still on supervised release. Journalist Brian Krebs notes that the group's current leader, a teenager going by "Ray" from ScatteredLapsussHunters, may have deliberately pinned ShinyHunters' recent breach of an FBI jobs site on van der Stap — an intrusion that reportedly leveraged a modified exploit for the Oracle PeopleSoft flaw CVE-2026-35273, previously abused as a zero-day. Van der Stap, who now works in offensive security, is also not believed to be the voice behind the social-engineering call that led to the massive Odido data breach affecting 6.2 million customers; he is due in court the same day as the police statement.

Apple has released security patches addressing CVE-2026-86950, a zero-day vulnerability in the Core Graphics framework that was actively abused in a highly targeted attack against specific individuals running versions prior to iOS 27. The flaw, discovered by Meta Product Security, is an out-of-bounds write bug that lets attackers achieve arbitrary code execution once a device processes a specially crafted malicious file through Core Graphics, the component responsible for rendering graphics, images and PDFs on Apple's platforms. Fixes are included in iOS 26.7.1, iPadOS 26.7.1, macOS Tahoe 26.7.1 and macOS Sequoia 15.8.1, while Apple's newest releases—iOS 27.0.1, iPadOS 27.0.1 and macOS Golden Gate 27.0.1—appear unaffected. Given the real-world exploitation and the sensitivity of the bug, all users are urged to install the update immediately.

Dutch law enforcement has confirmed the arrest of a 24-year-old man from Amsterdam tied to an ongoing investigation into the ShinyHunters hacking collective. Police stated he is due to appear before the Rotterdam District Court on September 29, 2026, though officials have not released further case details. Journalists Brian Krebs and DataBreaches.Net have identified the suspect as Pepijn van der Stap, previously arrested in 2023 for data theft and extortion, who had worked at security firm Hadrian and volunteered with the Dutch Institute for Vulnerability Disclosure before later taking an offensive security role at Neo Security. The arrest follows ShinyHunters' recent claim of breaching the FBI's job application portal, apply.fbijobs.gov, allegedly exfiltrating large volumes of sensitive data. The group insists the intrusion was not extortion or financially motivated, framing it instead as a demonstration. Investigators now believe the attackers did not use a genuine zero-day in Oracle PeopleSoft as claimed, but instead bypassed web application firewall protections meant to block exploitation of CVE-2026-35273 using a URL-encoding evasion technique.

Apple has released patches for a zero-day bug in CoreGraphics, the graphics and text-rendering framework shared across iOS, iPadOS, macOS, watchOS and tvOS, after confirming it was abused in highly targeted real-world attacks. The flaw, an out-of-bounds write weakness found by Meta Product Security, can let an attacker execute arbitrary code simply by getting a device to process a maliciously crafted file. Apple says it has reports of exploitation against specific individuals via what it calls an extremely sophisticated attack affecting iOS versions before iOS 27. Impacted hardware spans iPhone 11 and newer, several iPad Pro, Air, mini and standard models, plus Macs on macOS Sequoia and Tahoe; fixes are now available in iOS/iPadOS 26.7.1, macOS Tahoe 26.7.1 and macOS Sequoia 15.8.1, and users are urged to update promptly even though the attacks appear narrowly targeted.

Security researchers at OX disclosed CVE-2026-93355, an authentication bypass in the open-source LLM gateway LiteLLM that lets an attacker take over any existing user's account, including a proxyadmin, with a single forged request. The flaw stems from LiteLLM's JWT handling: when a token's userid or ssouserid doesn't directly match an account, the proxy falls back to matching on the token's email claim without ever checking whether that email has actually been verified. Worse, a successful match not only grants the attacker the victim's session but silently and permanently rewrites the victim's ssouserid to the attacker's token subject, turning a one-time forged claim into lasting access to every API key, budget, and admin function behind the proxy. Rated CVSS 8.8, the issue affects all versions up to 1.100.1 and remains unpatched as of this disclosure despite being reported to the vendor back in May 2026; a working proof of concept is now public, so administrators are urged to enforce verified-email claims and pin identity to stable subject IDs as interim mitigations.

Apple has issued updates for iOS, iPadOS, and macOS to address a zero-day flaw that attackers may already be abusing in targeted operations. The bug, CVE-2026-86950, lives in the CoreGraphics component and is an out-of-bounds write that can trigger code execution when a device processes a maliciously crafted file, potentially through web content, email, or messaging apps, raising the possibility of zero-click attacks. Apple credits Meta's security team with reporting the issue and says it has indications the flaw was used in a highly sophisticated campaign against a small number of specific individuals running iOS versions prior to iOS 27. The fix is included in iOS and iPadOS 26.7.1, macOS Tahoe 26.7.1, and macOS Sequoia 15.8.1, while newer iOS 27 and macOS Golden Gate 27 builds appear unaffected; CISA has not yet listed the CVE in its Known Exploited Vulnerabilities catalog.

Palo Alto Networks' Unit 42 researchers released OperTraitor, an open-source LLM-based scanner that compares a Kubernetes operator's documented purpose against the RBAC permissions it actually requests, flagging components that claim far more access than they need. Running it against OperatorHub and local installs turned up widespread over-permissioning, including many abandoned or outdated listings still carrying wildcard cluster-admin-style rights. The most serious finding involved IBM's Turbonomic platform, where the Prometurbo operator's service account held an unnecessary ClusterRole granting cluster-wide read access to Kubernetes Secrets — tracked as CVE-2026-6389 with a CVSS score of 8.8. IBM has since confirmed and fixed the issue, properly scoping the operator's permissions, while researchers warn that as operators increasingly incorporate AI agents, such RBAC misconfigurations will become far more dangerous attack vectors.

Security researchers at Ridge Security have disclosed a zero-day flaw in TDengine, an open-source time-series database widely deployed across manufacturing, energy, automotive and IoT environments, with over 730,000 instances reportedly in use at companies including Siemens, McDonald's, Sinopec and NavInfo. Tracked as CVE-2026-42542 (CVSS 7.5), the bug is an integer-underflow issue in TDengine's pre-authentication message handling on its default RPC port 6030, allowing an attacker with network access to crash the server by sending just one malformed packet—no credentials or session required. The flaw affects versions 3.4.0.0 through 3.4.1.5 and has been fixed in 3.4.1.6; while no in-the-wild exploitation or public exploit code has surfaced yet, Ridge Security built its own working proof-of-concept and warns that reproducing the bug would take hours rather than weeks. Because losing the database can mean losing visibility into industrial equipment and telemetry, the company urges affected organizations to update promptly or at least limit exposure to port 6030.

Apple has pushed out fixes for a CoreGraphics bug, CVE-2026-86950, affecting older builds of iOS, iPadOS, and macOS, warning it may already have been abused in highly targeted attacks against specific individuals. The flaw is an out-of-bounds write that can trigger arbitrary code execution when a device opens a specially crafted file, and it was reported to Apple by Meta's Product Security team. Apple did not disclose how many victims were hit, whether any attacks succeeded, or when exploitation first began. The patch is included in iOS 26.7.1 and iPadOS 26.7.1 (iPhone 11 and newer, several iPad Pro/Air/mini models), macOS Tahoe 26.7.1, and macOS Sequoia 15.8.1, and follows a similar actively exploited dyld flaw (CVE-2026-20700) Apple patched back in February.