🚨 Veeam, Terraform MCP, and Django patched 11 flaws exposing credentials, crossing tenant boundaries, and enabling file writes.

Veeam fixed CVE-2026-58073, which can expose managed-agent credentials without authentication, and CVE-2026-58072, a file-write flaw that can lead to RCE.

HashiCorp patched CVE-2026-16498 and CVE-2026-16496, which can break tenant isolation in Terraform MCP Server, plus SSRF flaw CVE-2026-14869.

Django fixed CVE-2026-15307, a GeoDjango flaw that can write files and, on some setups, lead to code execution.

See what needs patching: https://thehackernews.com/2026/08/veeam-terraform-mcp-django-patch.html

⚠️ Kali365 turns Microsoft's real device login page into a phishing trap.

Victims enter an attacker-controlled code, which may give attackers continued access to Microsoft 365 email and files.

AnyRun records 80+ public sessions a week, with the US the main target.

Read the article: https://thehackernews.com/2026/08/kali365-weaponizes-microsoft.html

⚡ Paperclip AI flaws can turn a malicious agent import into commands running on the host.

One chain reaches exposed servers through default registration. Another crosses into a developer’s localhost through DNS rebinding.

A public Metasploit module automates the server-side attack.

See how both chains work ↓ https://thehackernews.com/2026/08/paperclip-ai-flaws-let-attackers-run.html

‼️ Cheap Claude access can expose every prompt.

Researchers found 6+ underground AI proxy services. Poison Claude offered models at 5–15% of official prices through pooled accounts funded with abused AWS credits, with 872 active users exposed.

As an intermediary, it could read or leak prompts, swap models, or vanish without warning.

Read ↓ https://thehackernews.com/2026/08/poison-claude-sells-discounted-claude.html

⚠️ OpenAI banned a Poipet-based ChatGPT network supporting investment, romance, gambling, and fake police scams.

The accounts built personas, translated messages, forged documents, and may have contacted hundreds of targets.

Inside its ping-zing-sting playbook: https://thehackernews.com/2026/08/openai-disrupts-poipet-scam-network.html

🚨 Same URL. Malware for Macs, decoys for scanners.

Microsoft tracked over 250 ClickFix domains using browser fingerprinting to decide who sees the fake GitHub download and who sees nothing suspicious.

Inside the cloaking system: https://thehackernews.com/2026/08/over-250-clickfix-domains-use-browser.html

⚡ Snowflake hacker pleads guilty.

Connor Moucka admitted his role in breaches of at least 165 customer accounts, exposing records tied to at least 100 million people. Prosecutors say he made at least $495,000 from ransoms and data sales.

How years-old credentials got them in: https://thehackernews.com/2026/08/snowflake-hacker-pleads-guilty-over.html

🚨 On-premise TeamCity is under attack.

CISA says attackers are exploiting CVE-2026-63077 in the wild. The 9.8-rated flaw can let unauthenticated attackers run OS commands, putting stored credentials, build artifacts, and downstream CI/CD pipelines at risk.

What is known so far 🠖 https://thehackernews.com/2026/08/cisa-flags-teamcity-cve-2026-63077-rce.html

🚨 Ransom Cartel’s creator gets 16 years in prison.

Maksim Silnikau did not carry out most attacks himself. He bought network access, supplied the ransomware, ranked affiliates, handled victim negotiations, and moved payments through crypto mixers.

DOJ says the operation hit at least 18 companies.

Read: https://thehackernews.com/2026/08/ransom-cartel-creator-gets-16-years-in.html

‼️ Chinese-made routers shipped with a root backdoor.

ENDLESSDOORS was found in at least 20 Zbtlink models. It starts at boot and can open a live root shell without authentication, even when the router is not exposed to the internet.

Details here: https://thehackernews.com/2026/08/chinese-made-zbtlink-routers-ship-with.html

🛑 The model never runs. Its tools still do.

Newly reported security flaws in AWS AgentCore, Google ADK, and Vercel AI SDK harnesses let forged or untrusted instructions reach agent tools.

In several paths, system prompts, filters, and model guardrails never see the request.

Here's how each bypass reaches the tool layer: https://thehackernews.com/2026/08/aws-google-and-vercel-patch-agent-flaws.html

🚨 Attackers turned a web SQL injection into Windows SYSTEM access.

They compiled the khunt toolkit inside the Oracle database, then ran OS commands without dropping an executable to disk, beyond where typical endpoint tools look.

Look inside the attack chain: https://thehackernews.com/2026/08/attackers-compile-khunt-inside-oracle.html

‼️ Apple’s iCloud Private Relay can expose the real IP address it is supposed to hide.


Researchers found that a website can deliberately link a Safari session to the device’s real IP address, even with Private Relay enabled and without user interaction.

Here’s how the leak works: https://thehackernews.com/2026/08/webkit-proxy-bypasses-can-expose-real.html

🔥 $5.7 million stolen via guessable wallet recovery phrases.

Coinspect traced the Ill Bloom drains to weak CryptoJS randomness used by 5 wallet apps and estimates thousands of users across #Bitcoin and EVM networks may be affected.

‼️ An app update won’t secure an affected phrase; users must replace it and move their funds.

See how the flaw worked: https://thehackernews.com/2026/08/cryptojs-weak-rng-behind-57-million-in.html

⚠️ One click can poison your AI’s memory.

Websites are hiding instructions inside “Ask AI” buttons that may tell ChatGPT, Claude, Gemini, or Grok to save a vendor’s domain as a trusted source, potentially biasing future answers without the user’s knowledge.

See how it works: https://thehackernews.com/2026/08/ai-recommendation-poisoning-how-ask-ai.html

🛑 4,407 Rockwell PLCs are sitting on the public internet.

Forescout found 22 in cities hit by recent U.S. water attacks, with 19 on the same carrier network. It could not confirm any were compromised.

See what the scans uncovered: https://thehackernews.com/2026/08/over-4400-rockwell-plcs-exposed-online.html

Security teams are busy. But are they actually reducing risk?

The State of Assumed Security from Horizon3 dives into exactly that question.

Across 750 security leaders and practitioners, there's a hard truth: Most teams measure activity, not resistance. But if you can’t prove an attacker would fail…you’re just guessing.

Read the full report here: https://thn.news/assumed-security-report

⚡ For some, Thursday is winding down. For others, it has just begun. For us, it means 30 more cybersecurity stories for this week’s #ThreatsDay Bulletin:

• Odysseus RCE
• Samsung takeover
• AI proxyjacking
• Chinese telecom footholds
• ClickOnce backdoors
• 846 npm packages
• Pre-prompt code execution
• XCSSET macOS upgrade
• LLM pentesting lessons
• Telegram takedown extortion
• Signal multi-phone support
• AI bug-report flood
• Active Directory takeover
• AI scam farms
• CTV ad fraud
• ScreenConnect phishing
• AI memory poisoning
• Guardrail bypass tactics
• Omada router hijacking
• Exposed access broker
• $20M bug bounties
• AI email insider threat
• Ransomware memory theft
• Shorter NuGet key lifetimes
• Ethereum-based malware C2
• AI prompt exfiltration
• Apple’s iCloud backdoor fight
• Device-code phishing surge
• Fake PDF ClickFix lures
• Coldcard phishing

All 30. One roundup. Read it here: https://thehackernews.com/2026/08/threatsday-odysseus-rce-samsung-one.html

‼️ ALERT - A new CPU attack lands after Spectre v2 defenses run.

MIT researchers disclose INTERRUPT INJECTION, which uses a precisely timed hardware interrupt to re-poison branch prediction. It triggered kernel mispredictions on AMD Zen 2 and two Intel CPUs. On Zen 2, unprivileged Linux code leaked kernel memory and read /etc/shadow.

How the timing gap works: https://thehackernews.com/2026/08/new-interrupt-injection-attack-can.html

🚨 Three new Cisco SD-WAN flaws just scored 9.9.

Cisco also patched a 9.8 IOS XE command injection bug. A separate IMC flaw with public PoC could let a low-privilege attacker gain root on a controller that can influence BIOS and Secure Boot.

Read: https://thehackernews.com/2026/08/cisco-patches-12-sd-wan-and-ios-xe.html

‼️ New Linux KVM escape flaw disclosed with public PoC.

Zapscape (CVE-2026-64561) could let an attacker with guest root break out of a nested VM and execute code as root on the host when nested virtualization is exposed to untrusted guests.

Learn which KVM setups are exposed: https://thehackernews.com/2026/08/new-zapscape-kvm-flaw-could-let.html