
Critical RCEs Hit React and Firewalls as Cloud Firms Tighten…
Coverage: 19 Dec 2025 – 21 Dec 2025 (UTC)
< view all daily briefs >Defensive guidance and platform guardrails led today’s agenda as organizations moved to contain high‑risk vulnerabilities and strengthen operational resilience. A maximum‑severity flaw in React Server Components drew swift exploitation and even ransomware use, as detailed by CSOonline. In parallel, a major provider outlined changes to configuration rollouts designed to avoid Internet‑scale outages, with Cloudflare committing to staged, health‑mediated deployments for all config changes.
AI operations with guardrails
A year‑end retrospective from Google Cloud emphasized security and governance alongside rapid AI iteration, noting new protection capabilities, responses to critical vulnerabilities, and controls for agentic systems. The review frames 2025 as a balance between scaling generative AI and embedding controls, training, and guidance to help enterprises operationalize AI while addressing risk.
Concretely, the team showcased hardening patterns for agentic workloads on Kubernetes: an GKE “Agent Sandbox” isolates untrusted, LLM‑generated code with gVisor, strict network policies, and restricted filesystems, while Pod Snapshots cut cold‑start delays from minutes to seconds by restoring pre‑warmed sandboxes. The approach targets a recurring risk: agents that generate and execute code can delete data, exfiltrate secrets, or move laterally if not strongly contained. The result is faster agent deployments that retain kernel‑ and network‑level isolation.
CrowdStrike detailed a measured path for AI SOC agents built on expert‑annotated data, reproducible benchmarks, human‑validated reinforcement learning, and strict guardrails. The company reports >98% decision accuracy for Detection Triage and Agentic Response agents when tested against Falcon Complete decisions, with bounded autonomy, source‑linked explanations, and continuous monitoring to resist prompt injection, poisoning, and evasion. Why it matters: pairing isolation at runtime with auditable, governed agents aims to reduce toil without expanding blast radius.
Advisories and active exploitation
Attackers are exploiting CVE‑2025‑14733 in WatchGuard Firebox, prompting CISA to add the flaw to its Known Exploited Vulnerabilities Catalog under BOD 22‑01. The entry underscores elevated risk from out‑of‑bounds write issues and requires Federal Civilian Executive Branch agencies to prioritize remediation. Organizations beyond the federal enterprise are strongly urged to treat KEV entries as high priority and to monitor for indicators of compromise while completing fixes. In web stacks, the React2Shell issue in React Server Components has been used for rapid ransomware deployment, underscoring how client‑server communication logic can become a high‑value attack surface when unpatched; immediate version verification and forensic review are advised.
Elsewhere in the management plane, CSOonline reports a remotely exploitable, unauthenticated RCE in HPE OneView (CVE‑2025‑37164) affecting versions 5.20 through 10.20. HPE has issued hotfixes, with no workarounds; network‑restricting the management interface is recommended until patched, and some upgrades require reapplying the hotfix. For secure email infrastructure, CSOonline confirms a zero‑day in AsyncOS affecting Cisco Secure Email Gateway and related components when spam quarantine is enabled; a patch is not yet available, and confirmed compromises require wiping and rebuilding appliances while tightening management access. On edge devices, BleepingComputer highlights more than 25,000 internet‑facing Fortinet systems responding with a FortiCloud SSO fingerprint amid active exploitation of authentication bypass CVEs, with attackers forging SSO events to gain admin access and exfiltrate configurations; CISA has directed federal agencies to remediate under a binding directive.
Cloud resilience and operations
Operational controls continue to expand in managed services. A restart API for Timestream for InfluxDB, announced by AWS, enables administrators and automation to trigger database restarts for resilience testing and faster mitigation without support tickets. Separately, AWS expanded Glue zero‑ETL for self‑managed database sources to seven more regions, offering automated, ongoing replication into Redshift via a no‑code setup that reduces ingestion engineering overhead. These capabilities bring more lifecycle and data‑movement controls within reach of platform teams seeking agility and fault‑injection coverage.
Following two high‑severity outages tied to instantaneous global configuration pushes, Cloudflare described a “Fail Small” plan to impose staged, health‑mediated rollouts on all config changes, harden failure modes to safe defaults, and fix emergency access flows by the end of Q1. The company’s Code Orange post emphasizes closing the gap between binary releases—already staged with automated rollback—and configuration updates that previously propagated in seconds. Why it matters: applying production‑grade gates to configuration attempts to reduce blast radius when signatures, classifiers, or policies misbehave.
State‑linked activity
Denmark’s defence intelligence service attributed destructive and disruptive campaigns to Russian‑aligned hacktivists, naming Z‑Pentest for a water utility incident and NoName057(16) for election‑period DDoS, as reported by Infosecurity. The statement aligns with a multinational advisory enumerating tactics used by pro‑Russian groups and reinforces coordinated attribution and information‑sharing to protect critical services and democratic processes.
In separate research, the Iranian Infy group resurfaced with updated Foudre and Tonnerre malware, a domain generation algorithm for resilient command‑and‑control, and selective Telegram‑based signaling, according to The Hacker News. Delivery shifted from macro‑laced documents to embedded executables, and the toolset includes RSA‑signed validation files and structured C2 directories. The findings point to sustained investment in credential theft and data exfiltration capabilities across multiple regions.