Cyware at FS-ISAC Summit 2026
Understand Where You Are on the CTI Maturity Curve
Daily Threat Briefing
Diamond Trail

Cyware Daily Threat Intelligence - October 18, 2026

4 min read
Share this article
shutterstock 2176637153

Summary

Attackers are probing the boundaries of code execution with a critical flaw in SGLang, where a single crafted filter expression can let them run arbitrary Python code and seize full control of exposed systems. Cyware highlights that this vulnerability, tracked as CVE-2026-111391, carries a CVSS v4.0 score of 9.2 and puts the scheduler process at risk through the /dumper/configure endpoint.

Drupal administrators face a race against time as two remote code execution bugs—CVE-2026-108097 and CVE-2026-108099—threaten to turn vulnerable websites into attacker-controlled assets. The urgency is underscored by Drupal's warning that exploits could emerge within hours, pushing defenders to patch or upgrade to Drupal 10 or 11 before opportunistic scans begin.

Researchers have uncovered a sweeping data exposure risk in MCP servers, where 787 instances were cataloged and 150 confirmed vulnerable to a flaw dubbed 'filesystem capability laundering.' By bypassing path-validation checks, attackers can retrieve sensitive host files, prompting 13 security advisories and two CVEs for Anthropic’s Git MCP Server.

CTA_1_LI_Thursday_Sunday_DTI_After_Overview

Top Vulnerabilities Reported in Last 24 hours

SGLang bug enables remote Python execution

A critical code-injection flaw in SGLang (CVE-2026-111391, CVSS v4.0 9.2; CVSS v3.1 8.1) can let an unauthenticated remote attacker run arbitrary Python code and fully compromise affected systems. The risk comes from how the eval() call is used inside the _evaluate_filter() method, which can be tricked with crafted filter expressions to execute attacker-supplied code. Attackers can target the /dumper/configure endpoint when DUMPER_SERVER_PORT is set, putting the scheduler process directly in reach. No active exploitation has been reported in the provided source. A fix or update is referenced as available from the vendor; organizations also cite input validation and monitoring for suspicious activity to reduce exposure.

Drupal warns of takeover-grade RCE chain

Drupal issued an urgent warning that two critical remote code execution issues could enable attackers to take over vulnerable Drupal websites when combined with other modules: CVE-2026-108097 (Field UI) and CVE-2026-108099 (Entity List). While the flaws are described as not directly exploitable in Drupal core alone, the practical outcome is the same for affected sites: attackers could gain the ability to run code and seize control. No known active exploits have been reported yet, but Drupal warned exploits could appear within hours or days. The advisory is framed as a race condition for defenders because public attention can quickly translate into opportunistic scanning. Patches are available for supported versions, and Drupal recommends moving to the latest Drupal 10 or 11 (with Drupal 8 and 9 end-of-life).

MCP servers expose host files at scale

Researchers describing a weakness they call “filesystem capability laundering” report that some MCP servers can be manipulated into handing over sensitive host files outside the access boundaries they claim to enforce. The team used a purpose-built tool, MCP-FSD, to catalog servers, infer least-privilege “contracts,” and test whether path-validation checks could be bypassed in real deployments. In their study, they identified 787 affected servers and confirmed 150 vulnerable instances under tested sandbox policies, turning what should be a controlled integration into a data-exposure pathway. Exploitation has been confirmed in those instances, meaning attackers can potentially retrieve sensitive files by crafting arguments that slip past declared scope. Coordinated disclosure led to 13 security advisories and two CVEs in Anthropic’s Git MCP Server, with fixes addressed through those advisories.

CTA_2_LI_Thursday_Sunday_DTI_Before_FAQ

Frequently Asked Questions

  1. What is CVE-2026-111391? A critical code-injection flaw in SGLang (CVE-2026-111391, CVSS v4.0 9.2; CVSS v3.1 8.1) can let an unauthenticated remote attacker run arbitrary Python code and fully compromise affected systems. The risk comes from how the eval() call is used inside the _evaluate_filter() method, which can be tricked with crafted filter expressions to execute attacker-supplied code.

  2. What is CVE-2026-108097? Drupal issued an urgent warning that two critical remote code execution issues could enable attackers to take over vulnerable Drupal websites when combined with other modules: CVE-2026-108097 (Field UI) and CVE-2026-108099 (Entity List). While the flaws are described as not directly exploitable in Drupal core alone, the practical outcome is the same for affected sites: attackers could gain the ability to run code and seize control.

  3. What is MCP-FSD? Researchers describing a weakness they call “filesystem capability laundering” report that some MCP servers can be manipulated into handing over sensitive host files outside the access boundaries they claim to enforce. The team used a purpose-built tool, MCP-FSD, to catalog servers, infer least-privilege “contracts,” and test whether path-validation checks could be bypassed in real deployments.

Share this article

Discover Related Resources