‼️ CSS inside webmail can capture passwords and take over accounts.

New research demos also show CSS/HTML attack paths across Gmail, Fastmail, Proton Mail, Yahoo Mail, and AOL Mail that can leak tokens, hijack UI actions, or lead to account takeover.

See what email CSS can now do: https://thehackernews.com/2026/08/new-css-attacks-can-break-webmail.html

🛑 Atlassian's AI assistant Rovo can be tricked into sending Jira and Confluence data to attackers.

Two prompt-injection paths can make Rovo pull data the signed-in user can access and send it to an attacker-controlled server.

See how both chains work: https://thehackernews.com/2026/08/atlassian-rovo-can-be-tricked-into.html

⚡ OpenAI hits pause on some work involving Astra, its next AI model.

Internal tests were strong enough that OpenAI says it cannot rule out “Critical” cyber capabilities, prompting tighter security controls.

Here's what triggered the pause: https://thehackernews.com/2026/08/openais-next-ai-model-astra-shows-cyber.html

🚨 Malicious VS Code extensions steal wallet data and developer secrets.

Solidity Pro targets crypto wallet vaults, API keys, SSH keys, and source-control tokens. Some versions wait hours or days before activating, after quick automated scans have already moved on.

Read: https://thehackernews.com/2026/08/solidity-pro-vs-code-extensions-steal.html

🚨 Head Mare poisons TrueConf installers with PhantomCore malware.

The group exploits two flaws in unpatched TrueConf servers to gain SYSTEM privileges, then replaces the legitimate client distribution with an infected copy.

Full attack chain: https://thehackernews.com/2026/08/head-mare-exploits-trueconf-flaws-to.html

🔥 10–50× more code. Security still moves at human speed.

AI can multiply dependencies, vulnerabilities, and fixes faster than teams can review them. More scanning can just mean a bigger backlog.

Learn how to secure AI-speed development without slowing it down.

Watch the webinar: https://thehackernews.com/2026/08/shipping-1050-more-code-watch-this.html

‼️ New research shows passkey protections can be bypassed without breaking FIDO2 cryptography.

Three demonstrated paths:

• Replay Windows-exposed assertions against Entra ID
• Recover synced passkey private keys from Google Password Manager
• Use Windows Hello keys from a compromised session

See how each works: https://thehackernews.com/2026/08/new-passkey-attacks-can-recover-synced.html

⚠️ Phishing can succeed in under a minute.

Median click: 21 seconds. Data entry: 28 seconds later.

Blocking the email isn’t enough if the attacker’s domains and links stay live.

See what email security is still missing: https://thehackernews.com/expert-insights/2026/08/the-blind-spot-in-modern-email-security.html

🚨 Kimsuky is moving AI onto its own servers.

The North Korean espionage group is running or configuring Ollama, GPT4All, and Msty locally, alongside RAG, coding, and transcription tools. The stack has not been seen used against a victim yet.

See what it found: https://thehackernews.com/2026/08/kimsuky-builds-offline-ai-stack-that.html

🛑 Rogue AI • Metabase 0-day • Spectre bypass • Webmail attacks • Router backdoors • MCP supply-chain malware • 440 poisoned packages • AI token jacking • Device-code phishing • $30M crypto attacks • 26 ransomware hits a day.

That was just one week.

Catch up with the latest Monday Cybersecurity Recap - https://thehackernews.com/2026/08/weekly-recap-ai-goes-rogue-metabase-0.html

🚨 China-linked Storm-1175 deploys previously undocumented StormEncryptor ransomware.

Microsoft says the group has shifted from Medusa and likely used N-able N-central CVE-2026-18577 for initial access. Storm-1175 has been observed moving from initial access to data exfiltration and ransomware deployment within days.

Inside the campaign → https://thehackernews.com/2026/08/china-linked-hackers-deploy-new.html

🚨 ALERT - Poisoned #WordPress plugin JSON creates rogue admins and web shells.

Seven BdThemes plugins were disabled after attackers gained write access to vendor-hosted data fetched inside wp-admin.

Here's how the backdoor was planted: https://thehackernews.com/2026/08/bdthemes-supply-chain-attack-poisons.html

‼️ Hackers shut down a turbine at a Polish CHP plant by crossing a private cellular network.

They pivoted from a compromised wind farm through the grid operator’s private APN into the plant’s OT network, then put Siemens PLCs into STOP mode.

CERT says it’s the first real-world attack it has seen using this APN route.

See the full incident explained → https://thehackernews.com/2026/08/hackers-breach-polish-power-plant.html

⚠️ Gunra ransomware breaches networks, steals data, and destroys backups.

Attacks have exploited Fortinet FortiOS/FortiProxy and Schneider Electric PowerLogic P5 flaws for initial access, before encrypting databases, NAS systems, and other key assets.

How Gunra gets in and spreads: https://thehackernews.com/2026/08/gunra-ransomware-exploits-fortinet-and.html

🛑 Malicious MCP servers can make AI coding agents exfiltrate SSH keys, .env secrets, source code, and customer data.

The attack, dubbed “GhostSplice,” splits a request across MCP channels so no single fragment looks overtly malicious. In the researchers’ tests, splitting the request in two raised average compliance across 11 API-tested models from 42% to 82%.

See how it works: https://thehackernews.com/2026/08/malicious-mcp-servers-can-split.html

⚡ Windows Plug and Play can be abused to gain SYSTEM.

Researchers used emulated USB devices to make Windows fetch signed vendor software, then chained package flaws to SYSTEM on Windows 11.

It can also work over RDP when USB redirection is enabled.

See how Plug And Pwn works: https://thehackernews.com/2026/08/researchers-turn-usb-auto-install-into.html

[Video] AI agents are keeping access long after they're needed.

There are 109 non-human identities for every human. 82% of organizations have found unapproved shadow AI, and only 21% have a proper way to shut agents down.

See how identity controls the full AI agent lifecycle: https://thehackernews.com/videos/2026/08/ai-agent-lifecycle-risks.html

🚨 Fake crypto startup hires three suspected North Korean IT workers.

Researchers created Ballena Azul, interviewed the candidates, signed them as employees, and gave them work VMs. No exploit was needed. The access came through the hiring process.

See how they got hired: https://thehackernews.com/2026/08/researchers-built-fake-crypto-startup.html

‼️ A malicious SIM can take over the modem from inside the device.

Researchers found 9 of 26 tested phones and cellular modules accept RUN AT commands from the SIM, including 6 of the 8 modules. On one commercial EV charger, they chained the interface to code execution.

Here's how the attack works - https://thehackernews.com/2026/08/a-malicious-sim-card-can-run-attacker.html

🚨 Mozilla revoked the Firefox and Thunderbird Linux signing key after an unencrypted copy was accidentally committed to a private repo.

No unauthorized access was found, but older downloads can stop verifying and some Firefox RPM updates may fail.

Read: https://thehackernews.com/2026/08/mozilla-revokes-firefox-and-thunderbird.html

🔥 OpenAI just released a more cyber-permissive GPT-5.6.

GPT-5.6-Cyber is built for vulnerability research, penetration testing, incident response, and exploit development, with reduced refusals for some higher-risk dual-use tasks.

Read the full story: https://thehackernews.com/2026/08/openai-launches-gpt-56-cyber-with.html