‼️ A malicious webpage could poison the AI running on your own machine.


A NemoClaw weakness can expose Ollama to DNS rebinding, letting an attacker alter its chat template with instructions that persist across conversations.
Read how it works - https://thehackernews.com/2026/08/a-malicious-webpage-could-poison-your.html

48% of Mirage2FA-targeted emails were potentially compromised.

ANYRUN uncovered 9,000+ potential compromise events involving password and cookie theft, SSO logins, and 2FA bypass. Activity is potentially linked to 4,532 organization email domains.

See how Mirage2FA works - https://thehackernews.com/2026/08/mirage2fa-surge-hits-4500-us-and-eu.html

UPDATE - Apple fixed the iCloud Private Relay IP leak in iOS 26.6.1 and macOS 26.6.2.

The WebKit issue could bypass proxies and expose users’ real IPs. The fix was not in the iOS 26.6.1 beta.

See the update: https://thehackernews.com/2026/08/webkit-proxy-bypasses-can-expose-real.html

⚡ U.S. sanctions Iran-linked hackers tied to critical infrastructure breaches.

Three alleged Mabna Institute members are accused of stealing data from U.S. energy, defense, healthcare, IT, and financial organizations.

TRM Labs linked 30 wallets to $16.8 million.

Who they hit and where the money went: https://thehackernews.com/2026/08/us-sanctions-iran-linked-hackers-behind.html

🚨 Attackers are exploiting a critical Gitea RCE flaw.

CVE-2026-60004 can let an external attacker gain repo write access via open registration and run shell commands as the Gitea service account.

One reported attack dropped cryptocurrency-miner-like malware.

Read: https://thehackernews.com/2026/08/critical-gitea-rce-actively-exploited.html

58 arrested in INTERPOL crackdown on West African cyber-enabled fraud.

Operation Jackal IV identified 263 suspects across 22 countries. One Romanian investment scam is estimated to have stolen and laundered €143 million globally.

Inside the operation: https://thehackernews.com/2026/08/interpol-operation-jackal-iv-arrests-58.html

⚠️ Russian operators used ChatGPT to run an influence operation.

OpenAI banned the accounts after finding they used VPNs to bypass restrictions and generated posts promoting the International Burke Institute across Substack, Telegram, X, Facebook and LinkedIn.

What OpenAI uncovered: https://thehackernews.com/2026/08/openai-bans-russian-chatgpt-accounts.html

‼️ Nobody asked #Claude to cancel another gym member's booking.

Aikido rebuilt the Australian gym app and tested Opus 4.6 on #OpenClaw. The model bypassed its browser-only 7-day booking limit in 9 of 10 runs.

In two runs, it also canceled another member’s reservation.

Read: https://thehackernews.com/2026/08/claude-opus-46-bypasses-gym-booking.html

🚨 Unpatched Kaltura (a video management and streaming platform) flaws expose server files and could enable RCE.

The mwEmbed bugs need no authentication. Both stem from unsafe deserialization, and CERT/CC says no fix is available.

Read details here: https://thehackernews.com/2026/08/unpatched-kaltura-mwembed-flaws-could.html

Agentic SOCs can investigate alerts before analysts touch them.

AI agents can validate detections, test hypotheses against network telemetry, and return evidence-backed cases in seconds or minutes.

The shift: investigate first, escalate to humans when the evidence warrants it.

How the model works: https://thehackernews.com/2026/08/imagine-soc-without-queue-from-alert.html

🛑 CISA red teams compromised two critical infrastructure organizations. One SOC detected nothing.

The other caught the initial phishing and isolated affected workstations within 2–20 minutes.

Learn why one SOC saw it and the other didn’t: https://thehackernews.com/2026/08/cisa-red-team-compromised-two-critical.html

🔥 FBI shuts down Chinese hacking platforms tied to intrusions at NASA, the Fed, and U.S. Senate.

QScan exploited vulnerable IoT devices, while QTRouter hid attack traffic behind compromised devices, commercial proxies, and VPSs.

How it worked: https://thehackernews.com/2026/08/fbi-disrupts-china-linked-qtfy.html

⚠️ CISA added six exploited flaws to KEV, spanning NetScaler, Linux, SQL Server, Red Hat, and AjaxPro.

NetScaler CVE-2026-8452 stands out: web shells were dropped, with 36 exploitation attempts from 12 IPs in 12 days.

CISA also says injection flaws dominate recent CVEs, while AI is being used to automate exploitation.

Read: https://thehackernews.com/2026/08/cisa-adds-six-exploited-flaws-to-kev.html

‼️ New GPUThor attack defeats ECC on NVIDIA’s RTX A6000 and demonstrates host privilege escalation to root.

The Rowhammer technique uses multi-bit memory errors to corrupt GPU page tables, but requires an unprivileged CUDA kernel on the target GPU.

How the attack works: https://thehackernews.com/2026/08/gputhor-rowhammer-defeats-ecc-on-nvidia.html

🚨 New GoCaracal malware can retrieve fallback C2 addresses from Ethereum smart contracts.

Seen in a June intrusion at a Venezuelan communications organization, GoCaracal can read a replacement address from Ethereum after repeated primary C2 failures, then retry normal off-chain communications.

How the fallback works: https://thehackernews.com/2026/08/gocaracal-malware-uses-ethereum-smart.html

🚨 Spark RAT targets Cambodia using a vulnerable driver to kill security processes.

The campaign abuses CVE-2026-36425 in an OPSWAT driver to terminate processes tied to Microsoft Defender, Huorong, and Tencent PC Manager before deploying the RAT.

How the attack chain works: https://thehackernews.com/2026/08/spark-rat-targets-cambodia-abuses.html

A single public GitHub issue was enough to take over a Google Cloud project.

New research from Pillar Security details how an unauthenticated attacker could move from a comment on Google's Gemini CLI to Editor-level access inside a GCP project.

Prompt injection through the issue text reached an agent whose tool allowlist had gone inactive, which led to code execution on the CI runner, cloud credentials stored in plaintext, and an over-broad permission that allowed service-account impersonation up to Editor.

The permission traced back to Google's official setup script for Gemini CLI in GitHub Actions, so any team that followed the recommended setup could have inherited the same path. Google has since shipped fixes.

Research by Dan Lisichkin.

Learn more: https://thn.news/gcp-project-risk

🚨 A malicious Amazon Kiro workspace can leak sensitive local data.

After opening the workspace, sending any message to the agent can trigger repository-controlled instructions that send local data to an external server. Amazon fixed the flaw in Kiro 0.8.140.

How the flaw works: https://thehackernews.com/2026/08/amazon-kiro-prompt-injection-can.html

28% of security alerts go uninvestigated.

In a survey of 250+ security pros, 60% said a missed or ignored alert later became a serious issue. Meanwhile, 40% of teams use AI daily, and 72% of AI users report cutting investigation time by at least 25%.

What AI is changing inside the SOC: https://thehackernews.com/2026/08/what-data-says-about-ai-in-security.html

⚠️ Next.js patched two critical flaws that can enable unauthenticated RCE.

One affects Windows-hosted apps. The other requires AVIF optimization. No exploitation has been reported.

Details: https://thehackernews.com/2026/08/nextjs-patches-critical-avif-and.html