Eliran Emuna

Cybersecurity & AI 🛡️🤖 Breaking CVEs, threat intel & AI agents — verified, fast, actionable. Daily IOC drops ⬇️

[Eliran Emuna @EmunaEliran]

🛡️ Open Threat Intel IOC Feed — v1.0.0

🔎 IPs, domains, URLs and SHA256 / SHA1 / MD5 hashes
📊 Every indicator scored 0–100, with tiers
🔄 Auto-updated

Built for SOC, CTI and detection teams.

📦 https://t.co/siqUPxbNmR

[Eliran Emuna @EmunaEliran]

🚨 Two CVSS 10.0 flaws in Azure services
☁️ CVE-2026-83938: Azure Kubernetes Configuration Service, missing auth
🖥️ CVE-2026-110063: VM Inspector, link following
✅ Mitigated by Microsoft, no customer action, not exploited
🔗 https://t.co/29LHH7Rtjw
#Azure #CloudSecurity

[Eliran Emuna @EmunaEliran]

🔧 What's new in Ubuntu 26.10 (Oct 15): post-quantum-ready package signing, 100% Rust coreutils
🛡️ Sequoia PGP replaces GnuPG for APT checks
📦 Opt-in local speech-to-text; Entra ID MFA
Interim release, 9 months support: test or upgrade
https://t.co/SklLbExQ6T
#Ubuntu #Linux

[Eliran Emuna @EmunaEliran]

🚨 A robot dog whose GPT chat layer lets attackers run OS commands: Unitree Go2, CVE-2026-115995
⚠️ chat_go command injection, remote, no auth
📊 CVSS 3.1 9.6 · 4.0 9.4 (CNA)
🛡️ Not in KEV · no public PoC
✅ Fixed in 1.1.11.5
🔗 https://t.co/XCczeFHefA
#Robotics #IoT

[Eliran Emuna @EmunaEliran]

🧠 Google Cloud's Agent Compute Engine gives each AI agent its own hardware-isolated microVM

📦 Already powers Gemini; preview in the coming months
📊 One snapshot forks into thousands of copies for eval and RL

#AI #CloudSecurity
🔗 https://t.co/lMt3GnGxnv

[Eliran Emuna @EmunaEliran]

🤖 Anthropic hands Agent Skills, the open standard for reusable AI agent workflows, to the Linux Foundation's Agentic AI Foundation

📦 70+ agent platforms, incl. ChatGPT, Cursor, GitHub Copilot
📊 1M skills published, ~280M installs

#AI #OpenSource
🔗 https://t.co/aGuIY3ZAXR

[Eliran Emuna @EmunaEliran]

🤖 Moonshot AI's Kimi K3.1 Pro: open weights close to top proprietary coding agents

📊 Terminal-Bench 4.0: 65.7% vs GPT-6 Astra 58.2%, Claude Opus 5.5 66.4% (per Moonshot)
🧠 1M-token context
📦 Full weights by Nov 4

#AI #OpenSource
🔗 https://t.co/46iga5oiXc

[Eliran Emuna @EmunaEliran]

🚨 Oracle Alert: CVE-2026-114820, CVSS 3.1 9.1 (Oracle), Java SE/GraalVM JSSE, no auth
⚠️ Java SE 8u511, 11.0.33, 17.0.21, 21.0.13, 25.0.5, 27.0.1 + GraalVM
🛡️ Not in KEV
✅ Update to latest Java
🔗 https://t.co/C5Df7kugev
#Java #Oracle #CVE

[Eliran Emuna @EmunaEliran]

🚨 Brotli decoder RCE: CVE-2026-111497 (CVSS 4.0: 9.2)

⚠️ Out-of-bounds write via crafted input
📦 Brotli 1.0.3–1.2.0 (bundled in Node.js)
✅ Fixed: Brotli 1.0.10 · 1.1.1 · 1.2.1 | Node 22.23.4 · 24.22.1 · 26.12.1

🔗 https://t.co/k9hNP2ExFv

#CyberSecurity #NodeJS #CVE

[Eliran Emuna @EmunaEliran]

🚨 PHP: third RCE after PHP2Shell, CVE-2026-115560, no patch yet

⚠️ Hits only apps that switch session.serialize_handler to php at runtime
🛡️ Mitigation: session.upload_progress.enabled=0
🔧 Fix due in next PHP release

🔗 https://t.co/lpwTi4VRMc

#CyberSecurity #PHP #CVE

[Eliran Emuna @EmunaEliran]

🚨 Microsoft: treat AI agents as insiders

"Frontier models, regardless of their origin, cannot be trusted." (Hayete Gallot, EVP, Microsoft Security)

🛡️ Guardian Agents supervise every agent action before it executes
✅ Defender now discovers 60 local agent types, including Claude Code, OpenAI Codex and GitHub Copilot
✅ Dynamic permissions: task-scoped, time-bound, risk-adjusted
⚠️ Human approval required for high-impact actions

Takeaway: give every agent its own identity, least privilege and continuous monitoring.

🔗 https://t.co/BY5IQxvGJG

#AISecurity #CyberSecurity #AIAgents

[Eliran Emuna @EmunaEliran]

🛡️ NVIDIA OpenShell Reference Architecture: agent safety down to the silicon (GTC Berlin, Oct 21)

✅ Runtime: open-source OpenShell, with sandboxed agents and kernel-level policy enforcement
✅ Inference: Vera Rubin infrastructure for closed and open models
✅ Harness: Vera CPUs run agent harnesses and tool use
🚨 Enforcement: NVIDIA Sentry on BlueField-4 DPUs monitors agents out-of-band and quarantines one that leaves its boundary within milliseconds

⚠️ Key idea: enforcement runs outside the agent's reach, independent of the host.

🔗 https://t.co/FfUmoMw2z3

#AISecurity #NVIDIA #AIAgents

[Eliran Emuna @EmunaEliran]

🚨 CISA updates PHP2Shell alert (CVE-2026-109618 / CVE-2026-109619)

🛡️ New: SIGMA detection rule
⚠️ Can't patch, or on PHP 8.1 and older? Take PHP systems offline
✅ Fixed: 8.5.12 · 8.4.27 · 8.3.36 · 8.2.35

🔗 https://t.co/QYDtJDG8hU

#CyberSecurity #PHP2Shell #CVE

[Eliran Emuna @EmunaEliran]

🚨 PHP2Shell: two critical unauthenticated RCE flaws in PHP core

CVE-2026-109618 & CVE-2026-109619 | CVSS v4.0 9.2

⚠️ Affected: PHP serving web requests: mod_php (highest risk), PHP-FPM, php-cgi, FrankenPHP, LiteSpeed

🔗 Impact: A crafted request can corrupt PHP's memory before any application code runs, leading to code execution as the web worker.

📊 Status: Public PoCs and widespread exploitation attempts reported; CISA urges immediate patching.

✅ Fixed in: PHP 8.5.12 / 8.4.27 / 8.3.36 / 8.2.35

🛡️ Action:
• Patch internet-facing systems first
• Restart PHP workers, rebuild containers
• PHP ≤8.1: vendor backport or take offline
• Monitor for PHP processes spawning shells

🔗 https://t.co/VLftAL3N1n

#CyberSecurity #PHP #CVE

[Eliran Emuna @EmunaEliran]

🚨 SentinelOne launches Singularity ADR: Agentic Detection and Response

Announced Oct 21 at OneCon 2026, now in early access.

✅ Agent & MCP inventory: every agent, tool and MCP server, sanctioned or not
🛡️ In-line prevention: blocks prompt injection, jailbreaks, credential exfiltration and tool poisoning before execution
⚠️ Intent-drift detection: flags the gap between what an agent was asked to do and what it did
✅ Agent containment: stops one agent mid-action while the rest of the workflow keeps running

🔗 https://t.co/8qK7sfD4vx

#CyberSecurity #AISecurity #SentinelOne

[Eliran Emuna @EmunaEliran]

🚀 Start here:
1️⃣ Pick your platform's output in the repo
2️⃣ Follow the integration guide
3️⃣ Tune by score tier

📘 Guide: https://t.co/LCKSkx2ZPB
🌐 https://t.co/dKApUK2XlO

#ThreatIntel #IOC #BlueTeam

[Eliran Emuna @EmunaEliran]

🛡️ Open Threat Intel IOC Feed — v1.0.0

🔎 IPs, domains, URLs and SHA256 / SHA1 / MD5 hashes
📊 Every indicator scored 0–100, with tiers
🔄 Auto-updated

Built for SOC, CTI and detection teams.

📦 https://t.co/siqUPxbNmR

[Eliran Emuna @EmunaEliran]

🧩 Ready-made outputs:
Palo Alto EDL · FortiGate · Check Point · Cisco FMC · Microsoft Defender · CrowdStrike (API helper) · SentinelOne · Trend Vision One · Cortex XDR · Splunk · Microsoft Sentinel · Wazuh

➕ STIX 2.1 / CSV / TXT for everything else.

[Eliran Emuna @EmunaEliran]

🚨 Arista VeloCloud Orchestrator: CVSS 9.9 SQL injection, actively exploited (CVE-2026-114625)

CISA added it to the KEV catalog on Oct 23. On on-prem VCO, any authenticated user can gain operator privileges and masquerade as the internal "superuser operator", reaching functionality never meant to be remotely accessible. Full compromise of the orchestrator and the SD-WAN data it manages is possible.

⚠️ Affected (on-prem):
5.2.0 to 5.2.3.17 · 6.1.0 to 6.1.3.9 · 6.4.0 to 6.4.2.11

🛡️ Fixed in:
5.2.3.18 · 6.1.3.10 · 6.4.2.12

⚠️ Patch now?
✅ On-prem VCO: YES, today. CISA's federal deadline is Oct 26
✅ Hosted and Dedicated VCO: already patched by Arista
✅ Keep the VCO management interface off the internet and review operator-level activity and new accounts

🔗 https://t.co/rttZcK61nw

#CyberSecurity #CVE #Arista

[Eliran Emuna @EmunaEliran]

🚨 Nextcloud + Collabora built-in CODE Server: exploited unauthenticated RCE (CVE-2026-110447)

An OS command injection in the richdocumentscode app's proxy.php lets an unauthenticated request run commands as the web server user. CVSS v4.0 9.2. CISA added it to KEV on Oct 23 (federal due date: Oct 26). Users have reported webshells on affected Nextcloud servers.

⚠️ Affected:
Collabora Online - Built-in CODE Server (richdocumentscode) < 26.4.303
The attack works only while the built-in CODE server is not running (after a restart, upgrade or crash).

✅ Not affected:
Standalone Collabora Online (Docker or distro packages) and hosted Collabora

🛡️ Fix:
Upgrade richdocumentscode to 26.4.303+ or remove the app. Disabling it is NOT enough. On EOL Nextcloud ≤ 29, upgrade Nextcloud or remove the app.

⚠️ Patch now? YES, today, for any internet-facing Nextcloud with this app installed.

🔎 Check your version:
sudo -u www-data php occ app:list | grep richdocumentscode

🔎 Hunt: review access logs for requests to richdocumentscode/proxy.php and look for unexpected PHP files in the Nextcloud web root.

Note: this is a different flaw from CVE-2025-66208; that fix does not cover it.

🔗 https://t.co/8GhilWdYuy

#CyberSecurity #CVE #Nextcloud

[Eliran Emuna @EmunaEliran]

🤖 Hermes Agent v0.22.5 is out (Nous Research)

An open-source AI agent that builds reusable skills and persistent memory as it works.

✅ Learning loop: skills + memory that improve with use
✅ 20+ messaging platforms
✅ Scheduled, unattended automations
🛡️ Sandboxed execution backends + MCP tool support

⚠️ Running an agent with tool access? Use a sandboxed backend and least-privilege credentials.

🔗 https://t.co/P8BfkJI57a

#AIAgents #OpenSource #HermesAgent

[Eliran Emuna @EmunaEliran]

🚨 GitLab CE/EE: CVSS 10.0 unauthenticated RCE (CVE-2026-110264)

An integer overflow in GitLab's regular expression compiler can let an unauthenticated attacker execute code on the server. Every version since 7.0 is affected.

🛡️ Fixed in:
18.11.13 · 19.0.10 · 19.1.9 · 19.2.9 · 19.3.6 · 19.4.3 · 19.5.1

⚠️ Patch now?
✅ Self-managed, internet-facing: YES, today
✅ https://t.co/mXG8CX83pB: already patched
✅ Dedicated: covered from 19.3.5 / 19.4.2-internal0

🔎 Check your version:
head -1 /opt/gitlab/version-manifest.txt

Note: this is a different bug from CVE-2026-97229 (CI/CD regex, fixed Oct 12). Patching for that one does not cover this.

🔗 https://t.co/cQhzx2JtuL

#CyberSecurity #CVE #GitLab