Open-source tool Sage puts a security layer between AI agents and the OS - Help Net Security
The open-source tool Sage acts as a security layer between AI agents and the operating system, intercepting and verifying actions like shell commands and file writes. This Agent Detection & Response (ADR) tool, inspired by EDR, uses multiple detection layers including URL reputation, local heuristics, and package supply-chain checks, while keeping most data local to protect privacy.
Cybersecurity M&A Roundup: 42 Deals Announced in February 2026 - SecurityWeek
In February 2026, 42 cybersecurity mergers and acquisitions (M&A) were announced, including significant deals like Arctic Wolf acquiring Sevco Security and Booz Allen Hamilton acquiring Defy Security. Other key acquisitions involved Check Point, Endor Labs, Experian, Palo Alto Networks, Proofpoint, Quantum eMotion, Sophos, Varonis, and Zscaler, enhancing capabilities in areas such as AI security, attack surface management, and data protection.
How We Hacked McKinsey’s AI Platform — CodeWall.ai
Article claims: An autonomous offensive agent successfully hacked McKinsey’s AI platform, Lilli, within two hours by exploiting an unauthenticated SQL injection vulnerability, gaining read and write access to its production database. The breach exposed millions of chat messages, sensitive files, user accounts, and the platform’s proprietary knowledge base, highlighting the critical vulnerability of AI prompt layers.
ShinyHunters claims ongoing Salesforce Aura data theft attacks
The ShinyHunters group claims to be actively stealing data from Salesforce Experience Cloud platforms by exploiting misconfigurations that grant excessive permissions to guest users. Salesforce advises customers to audit and reduce guest user permissions to the minimum required, disable guest access to public APIs, and remove the API Enabled setting from the guest profile to mitigate these attacks.
Fake CleanMyMac Site Uses ClickFix Trick to Install SHub Stealer on macOS
A fake CleanMyMac website is using a ClickFix attack to trick macOS users into installing SHub Stealer, a malware that harvests passwords and targets cryptocurrency wallets. The malware avoids Russian devices and uses a fake password prompt to gain access to the macOS Keychain and steal sensitive information, including crypto wallet recovery phrases.
Russia forged new cyber weapons to attack Ukraine. Now they’re going international
Russia has deployed new cyber weapons, linked to the FSB’s Center 16 unit, to attack Ukraine and is now targeting international infrastructure, as evidenced by a recent attack on Poland’s solar grid. This escalation signifies a shift from espionage to destructive cyber operations, with the potential to disrupt critical systems globally.
OpenAI to acquire AI security platform Promptfoo - Help Net Security
OpenAI is acquiring the AI security platform Promptfoo to integrate its technology into OpenAI Frontier, a platform for building and operating AI coworkers. This acquisition aims to enhance security and safety testing, integrate evaluation into development workflows, and provide oversight and accountability for enterprises deploying AI systems.
Malicious npm Package Posing as OpenClaw Installer Deploys RAT, Steals macOS Credentials
A malicious npm package named “@openclaw-ai/openclawai” has been discovered that impersonates an OpenClaw installer to deploy a remote access trojan (RAT) and steal sensitive macOS credentials, including Keychain data, browser information, and cryptocurrency wallets. The package uses social engineering tactics, such as a fake installation interface and a convincing iCloud Keychain prompt, to trick users into revealing their system passwords, enabling sophisticated data theft and persistent remote control.
Ericsson US discloses data breach after service provider hack
Ericsson US has disclosed a data breach after one of its service providers was hacked, leading to the potential exposure of personal data belonging to an undisclosed number of employees and customers. The breach, which occurred between April 17-22, 2025, may have compromised sensitive information such as names, addresses, Social Security Numbers, and financial details, with Ericsson offering identity protection services to affected individuals.
The FBI is investigating suspicious cyber activity on an internal system containing sensitive information related to surveillance operations. The bureau is assessing the scope and impact of the incident, which involves sophisticated techniques exploiting network security controls, and has confirmed the cyber incident without providing further details.
China-Linked Hackers Use Malware Trio for Telecom Espionage
A China-linked cyberespionage group, identified as UAT-9244 and linked to Famous Sparrow and Tropic Trooper, has been targeting telecommunications providers in South America since 2024 using a trio of new malware: TernDoor (Windows backdoor), PeerTime (Linux backdoor), and BruteEntry (credential brute-forcing tool). These tools are designed for persistent access, data collection, and expanding network reach, with PeerTime uniquely using the BitTorrent protocol for command and control.
OpenAI Launches Codex Security that Discover, Validate and Patch Vulnerabilities
OpenAI has launched Codex Security, an AI agent designed to discover, validate, and patch vulnerabilities in codebases, aiming to replace noisy traditional security tools. The agent builds threat models, validates findings through proof-of-concept exploits, and generates patches, demonstrating significant reductions in alert noise and false positives during its beta phase. OpenAI is also offering free access to Codex Security for qualifying open-source maintainers through the Codex for OSS program.
Anthropic Finds 22 Firefox Vulnerabilities Using Claude Opus 4.6 AI Model
Anthropic’s Claude Opus 4.6 AI model identified 22 vulnerabilities in Firefox, including 14 high-severity ones, which have been addressed in Firefox 148. The AI model also demonstrated a limited ability to create exploits, highlighting that vulnerability discovery is more cost-effective than exploit creation.
Over 100 GitHub Repositories Distributing BoryptGrab Stealer - SecurityWeek
A new information stealer named BoryptGrab is being distributed through over 100 GitHub repositories, capable of harvesting browser and cryptocurrency wallet data, system information, and user files. Some variants also deploy the TunnesshClient backdoor, which uses an SSH tunnel for command-and-control communication.
Hackers Spread Fake Red Alert Rocket Alert App to Spy on Israeli Users
Researchers have identified a malicious version of the Red Alert rocket warning app targeting Israeli Android users, disguised as an update via fake SMS messages. This spyware steals sensitive data like GPS location, SMS messages, and contact lists while appearing to function normally.
Cognizant TriZetto Data Breach Exposes Health Information of 3.4 Million Patients
A significant data breach at Cognizant TriZetto Provider Solutions has exposed the sensitive health information of over 3.4 million patients. The breach, which involved unauthorized access to external systems, went undetected for over a year, leading to the exfiltration of personal identifiers and medical data, and prompting offers of credit monitoring and identity theft protection for affected individuals.
Microsoft: Hackers abusing AI at every stage of cyberattacks
Microsoft reports that hackers are increasingly leveraging AI across all stages of cyberattacks, from reconnaissance and malware creation to post-compromise activities. Threat actors use generative AI to accelerate operations, scale malicious activity, and lower technical barriers, with examples including the creation of fake identities and the generation of malicious code.
Termite ransomware breaches linked to ClickFix CastleRAT attacks
Threat actors known as Velvet Tempest are employing the ClickFix technique alongside legitimate Windows utilities to deploy DonutLoader malware and the CastleRAT backdoor. Researchers observed these actions, noting Velvet Tempest’s history with various ransomware strains, though Termite ransomware was not deployed in this specific observed intrusion.
OpenAI Codex Security Scanned 1.2 Million Commits and Found 10,561 High-Severity Issues
OpenAI has launched Codex Security, an AI-powered agent that scanned over 1.2 million commits and identified 10,561 high-severity issues, including vulnerabilities in projects like OpenSSH and PHP. The tool builds context about a project to find complex vulnerabilities, validates them, and proposes fixes to improve system security.
GitHub - matrixleons/evilwaf: evilwaf is a penetration testing tool designed to detect and bypass common Web Application Firewalls (WAFs). · GitHub
EvilWAF is a toolkit for Web Application Firewall (WAF) testing and bypass, functioning as a transparent MITM proxy. It features TCP and TLS fingerprint rotation, Tor IP rotation, and an Origin IP Hunter to discover the real server IP behind a WAF, enabling direct bypass.
Wikipedia hit by self-propagating JavaScript worm that vandalized pages
A self-propagating JavaScript worm vandalized pages on Meta-Wiki after a dormant script was executed, temporarily halting editing across Wikimedia projects. The worm modified user scripts and pages, but the Wikimedia Foundation has since restored the content and resolved the security issue.
Hackers Allegedly Selling Exploit for Windows Remote Desktop Services 0-Day Flaw
A threat actor is allegedly selling a zero-day exploit for a Windows Remote Desktop Services vulnerability (CVE-2026-21533) for $220,000, which allows for privilege escalation to gain local administrative control. This high-severity flaw impacts numerous Windows versions and requires immediate patching and potential disabling of Remote Desktop Services if not essential.
Recent Cisco Catalyst SD-WAN Vulnerability Now Widely Exploited - SecurityWeek
A Cisco Catalyst SD-WAN vulnerability (CVE-2026-20127), initially exploited as a zero-day, is now being widely exploited by threat actors. This escalating exploitation, now internet-wide, targets systems by bypassing authentication and escalating privileges, with a significant spike observed on March 4th. Experts warn that exposed systems should be considered compromised due to the mass and opportunistic nature of these attacks.
Massive GitHub malware operation spreads BoryptGrab stealer
A massive GitHub malware operation has been discovered spreading the BoryptGrab stealer through over 100 repositories, disguised as software tools and game cheats. This stealer targets sensitive data including browser credentials, cryptocurrency wallets, system information, and user files, with evidence suggesting Russian threat actors behind the campaign.
Hackers abuse .arpa DNS and ipv6 to evade phishing defenses
Hackers are exploiting the special-use .arpa domain and IPv6 reverse DNS to bypass phishing defenses by creating A records that point to phishing sites, making them harder to detect. This method leverages trusted internet infrastructure, obscuring the malicious activity and evading traditional security measures.