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]

🧵 3/3

hxxps://at[.]adwgamestore[.]org/ — Vidar — url
hxxps://c[.]aot88[.]com/ — Vidar — url
hxxps://c[.]aww88[.]net/ — Vidar — url
94[.]154[.]32[.]128 — Remcos — ip
94[.]154[.]32[.]69 — Remcos — ip
179[.]199[.]199[.]60 — VShell — ip
74[.]82[.]197[.]86 — Arista VCO exploitation CVE-2026-114625 — ip
38[.]54[.]50[.]151 — Arista VCO exploitation CVE-2026-114625 — ip

🔗 https://t.co/8S35SPwjxV

[Eliran Emuna @EmunaEliran]

🧪 Daily Threat Intel Drop #3

🚨 50 high-confidence IOCs · defanged · family · type

94[.]154[.]43[.]195 — Unknown malware — ip
0181cad18706699a580bdc030b44ddf4c9da6ea180b1f02217595ad50a13649e — AMOS — sha256
054e9f8e29c25fa2e0715932e581ca1f1797376c4cc8526f933ec4a84b264445 — AMOS — sha256
055681d35d9033885b79d8f1b5e2a5b84802467d7f13aa2841c4274d6c781db5 — AMOS — sha256
05b9730ddf48a48932316f4bb6c8b7b12cb76f2543dd1771f20efaa1739f7472 — AMOS — sha256
0bb8d03bbb250b63f0f8d7b16e258cd87aac7a2b4a683393940e4ad806f9c87b — AMOS — sha256
104[.]105[.]4[.]229 — Vidar — ip
110[.]41[.]77[.]122 — Cobalt Strike — ip
15abce52aaa4a15777d4634cc6bc0f5d5e4c6adeb6a791dc021f85d7597cff34 — AMOS — sha256
196aac46c392e8b80fee334d4e04ef457e08ca0f85dfc5f0111b829b0036b9ef — AMOS — sha256
1p289c94[.]betballl90[.]app — ClearFake — domain
204[.]168[.]137[.]221 — Vidar — ip
🧵 38 more below in replies

📥 Full CSV: https://t.co/gejyAN1z89
🔗 Firewall feeds: https://t.co/2ImQH5b1RE

#ThreatIntel #infosec #IOC

[Eliran Emuna @EmunaEliran]

🧵 2/3

c[.]aww88[.]net — Vidar — domain
church[.]springsoflife[.]org — StrelaStealer — domain
cinderfetch[.]cc — Unknown malware — domain
civil[.]rezendenetto[.]adv[.]br — StrelaStealer — domain
commithealthandwellness[.]com — StrelaStealer — domain
concreterepairny[.]com — StrelaStealer — domain
hxxps://104[.]105[.]4[.]229/ — Vidar — url
hxxps://135[.]181[.]105[.]214/ — Vidar — url
hxxps://167[.]233[.]150[.]198 — Vidar — url
hxxps://204[.]168[.]137[.]221/ — Vidar — url
hxxps://37[.]27[.]4[.]20/ — Vidar — url
hxxps://37[.]27[.]4[.]38/ — Vidar — url
hxxps://37[.]27[.]9[.]42/ — Vidar — url
hxxps://5[.]75[.]217[.]106/ — Vidar — url
hxxps://77[.]42[.]9[.]174/ — Vidar — url

[Eliran Emuna @EmunaEliran]

🐉 Kali Linux 2026.3 is out
🛠️ Build-scripts refresh: consistent across platforms, now buildable in Docker or Podman
🖥️ KDE Plasma 6.7 · kernel 6.19 to 7.1.5
🔍 New tools: dalfox, katana, rustscan, snaffler-ng, sdrangel
https://t.co/v5zQbbAo82
#KaliLinux #Linux #CyberSecurity

[Eliran Emuna @EmunaEliran]

🚨 Python ssl now rejects known-compromised CA certs: CVE-2026-115557 (CVSS 4.0: 9.4, PSF)
🛡️ New denylist applies regardless of SSLContext policy
✅ Fixed: 3.11.18 / 3.12.16 / 3.13.18
⚠️ Patch now; 3.14/3.15 not yet listed as fixed
🔗 https://t.co/n3pF82ByxY
#Python #TLS

[Eliran Emuna @EmunaEliran]

🛡️ OpenAI moves frontier model training offline

🔒 No internet access during frontier training; agents use an "offline world" of cached web content.
🧪 Red-team agents found microVM escapes and paths to the internet, now remediated.

🔗 https://t.co/yXFnJxFbKt

#AI #AISafety

[Eliran Emuna @EmunaEliran]

🚨 Yoosee GM8135S cameras: Secure Boot signature check is broken, CVE-2026-116116
⚠️ Firmware 21.00.01.36 · local, high privs
📊 CVSS 4.0 9.3 · 3.1 8.2 (CNA)
🛡️ Not in KEV · exploit public
❌ Vendor silent, no fix · isolate them
🔗 https://t.co/jlBaijFIVM
#IoT #InfoSec

[Eliran Emuna @EmunaEliran]

AI, explained · Ep. 1 🎬

Your AI agent just got smarter, and it's not a new model. It's a skill.

🤖 Anthropic donated Agent Skills to the Linux Foundation's AAIF as an open standard.
📊 70+ platforms · 1M skills · ~280M installs

🔗 https://t.co/aGuIY3Z38j

#AI #AIAgents

[Eliran Emuna @EmunaEliran]

3️⃣ Don't trust detector leaderboards
The best detector on BIPIA caught 2% of AgentDojo injections at a 1% false-positive rate.

➡️ Evaluate on your agent's own tool outputs, report detection at low FPR, audit the training data.
https://t.co/KaRnnFraMB

#AISecurity #PromptInjection

[Eliran Emuna @EmunaEliran]

🧪 A low prompt-injection success rate may not mean your agent is safe. Three new arXiv papers show how injection evaluations mislead. A short guide to testing your agent's injection defenses 👇

[Eliran Emuna @EmunaEliran]

2️⃣ Test the whole workflow
INFUSE: 19 LLMs, 40 long-horizon tasks, 56 attack scenarios. Per-model ASR ranged from 3.2% to 41.9%. Prompt-level hardening cut mean ASR from 58.2% to 14.3%, with model-dependent utility trade-offs.

➡️ Re-test per model.
https://t.co/20HYvfzAj4

[Eliran Emuna @EmunaEliran]

🟠 FastMCP (GHSA-7xgj-8392-5q4r): post-transform component authorization was skipped on direct calls. Affected: fastmcp and fastmcp-slim >=3.0.0 <3.4.10 and >=4.0.0 <4.1.2.

✅ Update both.
https://t.co/9iQYDrhdF8

#MCP #AISecurity

[Eliran Emuna @EmunaEliran]

🧩 MCP security week: 3 GitHub advisories in 2 core MCP tools. Check your versions.

🔴 MCP Inspector: 2 critical RCEs (CVSS 3.1: 9.6), fixed in 1.0.3 (21 Oct)
🟠 FastMCP: medium authz bypass (CVSS 3.1: 6.8), fixed in 3.4.10 / 4.1.2 (23 Oct)

Details 👇

[Eliran Emuna @EmunaEliran]

🔴 MCP Inspector (GHSA-6x67-4w4w-9f3w, GHSA-4fc4-wcmh-4g86): connecting to a malicious remote MCP server could lead to local command execution, via a token-stealing MCP App or an OAuth callback switching to stdio. Affected: 0.20.0 / 0.13.0 through 1.0.2.

[Eliran Emuna @EmunaEliran]

📌 Symantec's takeaway: AI services, model gateways and MCP servers should assume they will be found and tested soon after exposure.

Full report with IOCs and MITRE ATLAS mapping:
https://t.co/qzvWFgpZXi

#AISecurity #LLMjacking

[Eliran Emuna @EmunaEliran]

🤖 An exposed LLM endpoint is now an attack platform. Symantec captured the full prompt of "Skynet", an autonomous agent built to hijack unprotected AI models and run attacks for up to 100,000 steps, while the model's owner pays for every step.

🧵 What was caught 👇

[Eliran Emuna @EmunaEliran]

🛡️ Symantec's hardening checklist
• Auth on every LLM API, model server and MCP endpoint, incl. model-listing and health paths
• Rotate, scope AI keys; watch spend
• Restrict outbound traffic from AI hosts
• IMDSv2, hop limit 1
• Least privilege + human approval for agents