← All threat briefs

Daily Cybersecurity Brief — 2026-10-27

SECURITY // 2026-10-27

Daily Cybersecurity Brief — 2026-10-27

High-Signal Security Journalism

  • Krebs on Security — No new posts in the past 72 hours from Krebs on Security feed.

Threat Intelligence & Malware Analysis

  • Attackers hijack Dahua IP cameras in Ukraine using P2P access and auth weaknesses — CERT-UA warns of unauthorized remote access to Dahua video surveillance systems; attackers access video streams, change device configs, and create third-party accounts via P2P and authentication weaknesses. (Source: The Hacker News)

  • Hackers infect 953 online stores with scanner-evading JavaScript, mostly on Magento — Sansec uncovered a long-running campaign (active since July 2025) compromising 953 Magento-based e-commerce sites; malicious JS loader injected via CMS blocks steals payment data from shoppers. (Source: The Hacker News)

  • Check Point threat intelligence report: Onamae Desktop Cloud breach, SBI data leak, TeamUpdraft supply chain attack — Japanese cloud provider Onamae compromised, 26,000+ servers affected; State Bank of India exposes 3,897 customer records; WordPress backup tool TeamUpdraft breached, UpdraftCentral keys abused to install backdoors on WordPress sites; LiteLLM supply chain attack confirmed. (Source: Check Point Research)

  • Microsoft Teams IT support scams deploy malware and ransomware — Hundreds of incidents where threat actors use Teams' external chat feature to impersonate IT support, deliver malware via throwaway tenants and consistent social engineering patterns. (Source: The Hacker News)

  • Microsoft 365 token-replay attacks compromise 97 accounts at 79 organizations — Huntress observed AitM relay campaign (July–Sept 2026) capturing sign-in sessions and replaying stolen tokens; 497 successful sign-ins detected across targets. (Source: The Hacker News)

  • uCreator Android spyware steals Telegram 2FA passwords and clones WeChat accounts — Rentable spyware kit hides inside trojanized Telegram/WeChat apps distributed outside official stores; Telegram component matches Baohuo backdoor. (Source: The Hacker News)

  • Between alerts: What attackers do in the spaces ordinary detection misses — SOC alert overload leaves gaps; network detection and response (NDR) connects individual alerts into attack timelines, revealing intrusion progression that point alerts miss. (Source: The Hacker News)

  • Leaked LiteLLM master key lets attackers steal GitHub tokens and hijack release pipeline — Agent Platform master key accidentally published to public GitHub repo (May 2026) enabled threat actor to push malicious Python package (v1.107.0) stealing cloud credentials and AI service secrets; also covered in Check Point report. (Source: The Hacker News, Check Point Research)

Vulnerabilities & Exploits

  • CISA adds two PHP2Shell flaws to KEV catalog, confirming active exploitation — CVE-2026-109618 and CVE-2026-111497 (both CVSS 9.2) enable unauthenticated RCE on affected PHP deployments; CISA urges immediate patching. (Source: The Hacker News)

  • PHP issues second Windows security update in five days for Brotli RCE (CVE-2026-111497) — PHP rebuilds Windows packages for 8.2–8.5 fixing Brotli decoder flaw reachable via cURL auto-decompression; "Broccoli2Shell" affects Node.js, Bun, Python, Envoy, .NET, and other Brotli consumers. (Source: The Hacker News, Wiz Blog)

  • Two critical Wasmtime flaws allow WebAssembly sandbox escape — Wasmtime (WebAssembly runtime) vulnerabilities let malicious modules escape containment; one flaw ARM64-specific; fixed versions released. (Source: The Hacker News)

  • Unpatched Tailscale SSH flaws let hacked device reach root on other servers — Single compromised device on Tailscale network can pivot to root on reachable servers when SSH port forwarding enabled; no patch yet. (Source: The Hacker News)

  • Microsoft Authenticator Android vulnerability (CVE-2026-80097, CVSS 8.6) lets rogue apps steal work account tokens — Enclave researchers found any app could impersonate Teams/Outlook to obtain tokens granting full mailbox, SharePoint, Teams access; fixed in Authenticator 6.2606.4246+. (Source: Help Net Security)

  • 71% of vulnerability exploitation starts within a week of disclosure — Rapid7 Q3 2026 report: avg 3 days from disclosure to exploitation for high/critical vulns; 43 newly exploited vulns tracked, 34 requiring no auth/user interaction; state actors (N. Korea, Russia, Pakistan) targeting gov/defense/finance. (Source: Help Net Security)

  • Cross-request KV cache exposure in GLM-5.3 on Cloudflare Workers AI — SGLang config mismatch (1M token context vs 681K KV cache pool) caused 155 responses to leak other users' data (Aug 27–Sept 9); model taken offline, patched, restored same day; coordinated disclosure with SGLang. (Source: Cloudflare Blog)

AI/ML Security

  • AI coding agents fall for malicious instructions planted in Downloads folder (AGENTS.md attack) — Forever Security caught live attack: website drops AGENTS.md in ~/Downloads; when developer opens folder in Codex, agent installs malicious npm package (function-color) and exfiltrates login tokens; Codex ran attacker's commands in >90% of tests; Claude Code more resistant. Same technique documented by Dark Reading as "in-the-wild attack abusing AI agents' trust in markdown." (Source: Help Net Security, The Hacker News, Dark Reading)

  • So many AI security benchmarks, but none tell the whole story — Robert Lemos (Dark Reading): patchwork of benchmarks (MiST, cyber-eval, etc.) measure different variables; orgs must test models against their own code/workflows; Dream Group trained 8B-param MiST on 220K vuln records for cyber tasks. (Source: Dark Reading)

  • Google updates privacy policy: Gmail/Photos data shared with AI Search/Gemini may be used for targeted ads — Policy change discloses that personalized AI features in Search/Gemini can leverage cross-service activity for ad targeting; users can choose which connected services share data. (Source: The Hacker News)

  • Risks & rewards: How AI is redefining the CISO's role — Nvidia's Mark Weatherford (ex-DHS, ex-CISO California/Colorado) at ISC2: CISOs must balance AI competitive advantage with risk ownership; "speed is biggest challenge" — pen testing 4×/year no longer adequate, continuous testing needed. (Source: Dark Reading)

  • Several AI security market categories headed for bigger platforms — KuppingerCole's Jonathan Care: AI security startups mirror early fraud market; consolidation into platforms likely; Q3 funding to fewer, larger rounds; Israel's cyber venture share growing; five controls needed before agent deployment. (Source: Help Net Security)

  • LLM agent security needs more than guardrails — Concentric AI's Madhu Shashanka: securing agents depends on data controls, not model guardrails; covers prompt injection, connector data leakage, overprivileged agents, shadow AI. (Source: Help Net Security)

Supply Chain & Software Security

  • Gheist campaign: automated abuse of stolen GitHub tokens at scale — Since Sept 15, Wiz tracked >13,500 planted branches from 850+ compromised accounts; sync-check workflow base32-encodes and exfiltrates repo secrets (including environment-scoped prod creds) via GitHub API, deletes runs; Tool B variant adds timeout, environment matrix, per-repo payload tuning; 2/3 accounts now unavailable (likely GitHub response). (Source: Wiz Blog, The Hacker News, Dark Reading)

  • The Pwn Request Reckoning: what still works against GitHub Actions — GitHub's 2026 phased rollout (checkout@v7 fork blocks, read-only cache tokens, cache-mode GA, Workflow Execution Protections, Dec 2 default event policy) significantly raises bar but doesn't eliminate pull_request_target risk; private/internal repos and workflow_run unaffected; maintainers adding event policies to allow pull_request_target are exactly the workflows handling untrusted input with secrets. (Source: Wiz Blog)

  • Amutable Foundations: Linux with no SSH, no package manager, hardware-rooted attestation — Private preview: machines boot verified images, record measurements in TPM/CPU attestation, send signed reports to control plane; config in version control, changes only via new signed images; ties identity to measured state (deny creds to outdated machines); addresses EU Cyber Resilience Act gaps. (Source: Help Net Security)

Cloud & Infrastructure Security

  • Kali Linux 2026.3 released: kernel 7.1.5, containerized builds, NetHunter wardriving on TicWatch Pro 3 — Build scripts overhauled, build-in-container.sh for Docker/Podman; 126 package updates; KDE Plasma 6.7 per-screen virtual desktops; new tools (dalfox, katana, qsreplace, rustscan, sdrangel, snaffler-ng); NetHunter smartwatch wardriving, NFC Mifare cloning, WearHunter phone control. (Source: Help Net Security)

Policy & Regulation

  • Google says Gmail and Photos data shared with AI Search could be used for targeted ads — Updated privacy policy explicitly allows cross-service data sharing for ad personalization when users enable AI features in Search/Gemini. (Source: The Hacker News)

Law Enforcement & Takedowns

  • CERT-UA alert on Dahua IP camera compromises in Ukraine — State Special Communications Service confirms active exploitation of Dahua surveillance gear; P2P access and auth weaknesses exploited for unauthorized video access, config changes, account creation. (Source: The Hacker News)

Vendor Advisories & Patches

  • CISA KEV: PHP2Shell flaws (CVE-2026-109618, CVE-2026-111497) actively exploited — patch immediately — Both CVSS 9.2, unauthenticated RCE; added to Known Exploited Vulnerabilities catalog. (Source: The Hacker News)

  • Microsoft Authenticator Android CVE-2026-80097 patched (CVSS 8.6) — Update to 6.2606.4246+ via Play Store; MDM should enforce minimum version. (Source: Help Net Security)

  • PHP Windows packages rebuilt for Brotli RCE (CVE-2026-111497) — Second rebuild in five days for PHP 8.2–8.5; flaw in upstream Brotli C decoder. (Source: The Hacker News)

  • Wasmtime sandbox escape flaws fixed — Two critical vulns patched; one ARM64-specific. (Source: The Hacker News)

Reports & Analysis

  • Rapid7 Q3 2026 Threat Landscape: exploitation accelerating, AI in 8% of ops — 3-day avg disclosure-to-exploit; 71% exploited within week; 43 high/critical vulns; state actors (N. Korea Void Blizzard, Russia, Pakistan Op C-Major); US gov ransomware share up 45% (25 incidents); AI operational in 8% of tracked reporting. (Source: Help Net Security)

  • Forrester Email/Messaging/Collaboration Security Landscape Q4 2026: AI agents turning comms into instructions is top disruptor — Varonis takeaways: AI raises stakes both sides (phishing scale vs agent defense), social engineering crosses channels (email→chat→file→browser), trust is new attack surface (malicious intent hides in legitimate comms); defenses must verify sender identity, contextualize requests, block pre-engagement. (Source: Varonis Blog)

IoT & OT Security

  • Dahua IP camera compromise in Ukraine (CERT-UA alert) — See Threat Intelligence section above. (Source: The Hacker News)