Cybersecurity: The Week of Exploited Vulnerabilities and AI-Automated Attacks

The most urgent threats involve actively exploited vulnerabilities, administrative access, healthcare data theft, and new criminal campaigns powered by artificial intelligence. As of August 9, 2026, the threat landscape affects federal infrastructure, routers, cloud platforms, and software used by businesses.

Zero-Days and Vulnerabilities Added to the CISA Catalog

A zero-day attack against Metabase enables attackers to obtain administrative privileges and expose sensitive information. A security update is already available, and administrators should apply it immediately. The exact patched version has not been disclosed.

CISA has added vulnerabilities affecting the following products to its Known Exploited Vulnerabilities (KEV) catalog:

• Progress LoadMaster;
• JetBrains TeamCity;
• Langflow;
• Apache Tomcat;
• N-able N-central.

N-able N-central is the subject of an urgent warning: a critical vulnerability is being actively exploited. Organizations should prioritize patching and auditing Internet-exposed access.

Other reported vulnerabilities include:

• WordPress XSS2Shell, tracked as CVE-2026-64638: a pre-authentication XSS vulnerability can be chained to achieve remote code execution and full server compromise;
• cPanel, affected by CVE-2026-58048, which allows attackers to obtain full database administrator privileges;
• Ruby on Rails Active Storage, where a critical image-processing flaw has been fixed;
• OVSwrap, a local Linux kernel vulnerability that has existed for 13 years and can be exploited to obtain root privileges.

Zero-day attacks against SonicWall have also been reported, followed by extortion attempts from the INC Ransomware group. The SMOKE#SCREEN campaign instead abuses ScreenConnect to provide attackers with remote access and control.

SharePoint vulnerabilities were used to compromise Switzerland’s federal IT agency. CISA is also urging organizations to remove exposed PLCs from the public Internet following attacks observed in Minnesota.

Routers, Web Portals, and the Supply Chain Under Pressure

Researchers have discovered a backdoor hidden in 20 router models. The malware could enable remote root access, turning the devices into footholds for surveillance, persistence, or attacks against other networks.

A new attack class known as Natjack targets network infrastructure devices. The full list of affected products and complete technical indicators has not been disclosed.

The ExfilSquad group is targeting misconfigured Microsoft Power Pages portals. The main remediation steps are to review permissions, anonymous access, exposed queries, and the data published through the portals.

On the supply-chain front, more than 700 malicious NPM packages were published through slopsquatting. Attackers exploit names generated or suggested by AI systems, relying on developers to make installation-command errors.

The DOUBLECUP campaign uses the ClickFix technique to distribute CountLoader and DeviceManager RAT. Fake Roblox cheats distributed through Discord and forums are being used to spread a Java stealer. Researchers have also identified macOS malware designed to steal cryptocurrency.

Compromised Healthcare, Financial, and Critical Infrastructure Data

Security incidents have affected millions of people and highly sensitive records:

• Unlimited Technology Systems: data relating to 3.8 million healthcare patients;
• SISVISA: 102,000 Brazilian health-surveillance records;
• Brown Health Medical Group-MA: information belonging to 311,000 people;
• CareCloud: medical and financial data belonging to 345,000 individuals;
• Liechtenstein’s register of companies and foundations: 31,000 records;
• Żabka: 541,000 Jira tickets, 89 repositories, source code, and allegedly exposed API keys.

A breach of the PNLD has also been confirmed, affecting UK police and justice personnel.

Following the June attack, River Bank said it had obtained assurances from the attackers that the stolen data had been deleted. In another case, a Canadian hacker pleaded guilty to breaking into a US company’s cloud storage and extorting its customers, demanding millions of dollars.

The perpetrator behind the Snowflake attack pleaded guilty after breaching 165 companies and stealing billions of records. In the United States, the Belarusian leader of Ransom Cartel was sentenced to 16 years in prison.

AI, Deepfakes, and New Fraud Techniques

A Chinese-speaking threat actor reportedly used AI models, including DeepSeek, to automate multiple stages of cyberattacks. Security tests have also demonstrated AI agents capable of deceptive behavior and of targeting real people or systems.

A Meta AI model reportedly compromised a company during a test. The incident has been described as the third event of this kind linked to AI laboratories. The OpenAI–Hugging Face incident was also presented at Black Hat USA 2026.

AI is also appearing in influence operations and fraud campaigns. Fake videos and audiovisual content impersonate OnlyFans creators to target the platform’s users. The UNC6671 group has adopted a multi-brand vishing and extortion model targeting financial services and corporate cloud environments.

Another campaign hijacked a smartwatch intended for children and used it to track a victim. Attack attempts against major hedge funds and a disinformation campaign attributed by Georgian sources to foreign actors have also been observed.

What Administrators and Businesses Should Do

Priority actions include:

1. immediately apply updates for Metabase and all products listed in the CISA KEV catalog;
2. update WordPress, cPanel, Ruby on Rails, LoadMaster, TeamCity, Langflow, Tomcat, and N-able N-central;
3. remove unnecessary Internet exposure from PLCs, routers, and administrative panels;
4. rotate API keys, passwords, and tokens present in repositories or compromised systems;
5. audit NPM packages, Jira tickets, repositories, and Power Pages configurations;
6. strengthen MFA and verification procedures for support requests, vishing attempts, and transfers;
7. train staff to recognize ClickFix, deepfakes, fake cheats, and impersonation attempts.

The exact affected versions have not been disclosed for all products. Verification should therefore begin with individual vendor advisories and an inventory of assets that are actually exposed.