Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
LlamaFirewall is a framework designed to detect and mitigate AI-centric security risks. It consists of a set of scanners designed to detect various security risks. LlamaFirewall supports multiple layers of inputs and outputs, such as typical LLM chat and more advanced multi-step agentic operations.
Cy4Data secures both structured and unstructured data fields, records, and tables at rest, in transit, and when processed. The company also provides a key management system and real-time threat detection.
Evercookie is a JavaScript API that creates extremely persistent cookies in a browser to track a client's activity after they have removed cookies.
Tech giants including Microsoft, Cisco, and IBM have drafted the "OpenEoX" framework through OASIS to standardize end-of-life security notifications for products. The proposal creates a machine-readable format defining four key lifecycle points: General Availability, End of Sales, End of Security Support, and End of Life.
WhatsApp has launched a new feature called Advanced Chat Privacy. This setting can be enabled on chats to add an additional layer of privacy by preventing users from exporting chats, auto-downloading media to their phone, and using messages for AI features. This feature comes in the wake of privacy concerns caused by the always-on Meta AI features.
DARPA Director Rob McHenry believes that a combination of LLMs and formal methods for software development could virtually eliminate software vulnerabilities. Formal methods are a laborious process that uses mathematical models to prove the correctness of software. McHenry believes that given the success of the DARPA AI Cyber Challenge, LLMs could be used with formal methods to generate and validate the correctness of proofs automatically.
Michael Scheuer, a former Disney Menu Production Manager, received a three-year prison sentence and a $688,000 fine for hacking Disney's restaurant menu application and altering critical information to include offensive content, thus creating safety risks.
Kali Linux users must manually install a new signing key for the repository due to a key change, which prevents system updates until the new key is added.
Mandiant has noted that North Korean operatives have infiltrated many Fortune 500 companies, posing a significant threat to IT infrastructure and data security.
Sponsored
As the avalanche of AI-generated code transforms development practices, AppSec teams face unprecedented security challenges.
VeriSource Services, which provides employee benefit administrative services, is notifying 4 million individuals that it experienced a data breach in February 2024. The compromised data includes names, addresses, dates of birth, gender information, and Social Security numbers.
Threat intelligence researchers from Australian firm Dvuln say that 31k passwords belonging to Australian customers of the Big Four banks are being shared on Telegram and dark web sites. The researchers stated that the passwords are coming directly from infostealers installed on victims' devices, not a vulnerability or breach from the banks. Dvuln has stated that these bank accounts can be used to transfer funds or for money laundering.
Researchers have discovered "AirBorne" security flaws in Apple's AirPlay protocol that allow hackers on the same Wi-Fi network to hijack AirPlay-enabled devices. While Apple has patched vulnerabilities in its products, millions of third-party devices remain at risk, potentially creating persistent network access points for attackers.
A panel in RSA discussed the role of the CISO, highlighting how CISOs should secure personal liability insurance and golden parachutes when starting new jobs as protection against becoming scapegoats. They advised documenting everything, avoiding lawsuits against employers after whistleblowing to prevent industry blacklisting, and not trusting HR departments, which ultimately protect the company, not employees.
Retrieval Augmented Generation (RAG) can unexpectedly make LLMs less safe by bypassing built-in guardrails. When using RAG, models that normally refuse harmful queries often produce unsafe responses. Generic AI safety frameworks frequently fail to address domain-specific risks in industries such as financial services.
AWS default roles, such as AWSGlueServiceRole, often have overly broad access, like AmazonS3FullAccess, making service takeover easier for attackers. They can exploit these roles to manipulate various AWS services beyond S3 buckets, risking the entire environment. To mitigate risks, audit AWS roles and restrict excessive access, especially to S3 buckets.
Sponsored
Builders. Innovators. Exceptional security leaders. People who are ready to do more – to develop and execute a vision for the future. Join a team [11] where your expertise and passion are valued, and outperformance is generously rewarded. Apply to be a Staff Blockchain Security Architect [12] or check out all available positions [11]. ↗️
OpenAIPot is a deceptive OpenAI API gateway that acts as a honeypot for detecting unauthorized API usage.
AuthMind monitors the flow of activities for both human and non-human identities, aiming to provide organizations with a consolidated identity security posture management (ISPM) and threat detection and response (ITDR) solution.
Produce secure packages and containers with declarative configurations.
The bipartisan Take It Down Act, which will require social media companies to take down content flagged as nonconsensual (including AI-generated) sexual images, is heading to Trump's desk for signature. Some privacy activists are concerned that the takedown will be selectively enforced by the FTC and does not provide enough time for smaller sites to verify the claims correctly, which could be abused to hinder platforms' ability to operate effectively. The bill does not contain exemptions for end-to-end encrypted services such as private messaging and cloud storage.
Oregon's Environmental Quality Department faced a cyberattack by the ransomware group Rhysida. Officials have not confirmed data theft claims, including the theft of employee information, although services such as vehicle inspections were interrupted.
Jeffrey Bowie, CEO of the cybersecurity firm Veritaco, was arrested for allegedly installing malware on the computers of St. Anthony Hospital. Security cameras captured him accessing computers, where he installed software that took screenshots and transmitted them externally. Hospital staff caught him, and forensic analysis confirmed the breach.
Google reported 75 zero-day vulnerabilities in 2024, down from 98 in 2023, with 44% targeting enterprise products, particularly security software.
After a two-week downtime caused by a hack, 4chan is back online.
Microsoft will begin charging a subscription fee in July for Windows Server 2025 hotpatching, which enables security updates without requiring a restart.
Sponsored
If you're navigating Microsoft's MFA requirement for Azure admin access, you're not alone. But did you know that you can use Cisco Duo to meet this requirement [4] through Entra ID External Authentication Methods.
Juice jacking attacks are when an attacker hides a malicious device inside a USB charging cable that either runs malware or downloads files upon connecting to a target. Apple and Android have attempted to mitigate this risk by requiring users of phones to confirm that they want to allow access to their files. A new attack first connects as a USB input device and pairs as a Bluetooth keyboard. It then uses USB PD Data Role swap to connect as a USB host and accesses files by using the Bluetooth keyboard to accept prompts.
Hitachi Vantara, a subsidiary of Hitachi that provides data storage, infrastructure systems, cloud management, and ransomware recovery services, announced that it has proactively taken down some of its servers due to detecting a ransomware attack in its network. A source has informed that the attack was carried out by the Akira ransomware gang, which has stolen data and dropped ransom notes on affected systems.
A critical flaw in Commvault software allows hackers to remotely take full control of systems. Enterprises are advised to patch against CVE-2025-34028 to prevent immediate attacks.
Semantic scatter plots help uncover patterns in large datasets by grouping similar data points on a graph, allowing for a more precise visualization of relationships. Shadowpuppet utilizes these plots to analyze unstructured data, like leaked communications from cyber extortion groups. By leveraging embedding models and dimension reduction, Shadowpuppet offers an intuitive method for exploring and visualizing complex data in threat intelligence analysis.
A cybersecurity study used Markov Chain simulations to compare two defense strategies: a prevention-only approach versus a detection and response approach. In the prevention-only strategy, attackers had a 49% success probability with about 2 attempts. Adding a 10% detection probability reduced success rates to 26% and increased the number of attempts to 3.84. The simulation showed that detection capabilities are more effective than layered defenses, as detection success rates of 30% provide equivalent protection to three layers of 50%-effective preventive controls, highlighting the SOC as a powerful security force multiplier.
Postal codes play a vital role in cybersecurity by facilitating fraud detection through Address Verification Services and e-commerce analysis. To safeguard postal code data, organizations should implement strategies such as limiting internal access, encrypting data, utilizing secure APIs, and regularly updating datasets. As postal codes evolve into components of digital identity verification, ensuring their security and accuracy is crucial to protect users from cyber threats and privacy invasions.
MCP-Scan is a security tool for checking MCP servers for common vulnerabilities, prompt injections, and tool poisoning. It scans server configurations and tool descriptions to detect security issues. Users can run MCP-Scan using simple commands such as "uvx mcp-scan@latest" or "npx mcp-scan@latest."
Anvil Secure's HANAlyzer is an open-source tool that automates security checks on SAP HANA databases. It generates reports using SAP's security checklist to identify and address vulnerabilities efficiently. HANAlyzer is designed to be easily deployed on multiple platforms. It simplifies the assessment and security of SAP HANA environments.
Know Your Enemies (KYE) helps you check external access to your AWS account by analyzing IAM Role trust policies and S3 bucket policies. It identifies third-party vendors with access to your resources and supports defining trusted AWS accounts. By running the script, you can detect IAM roles vulnerable to the confused deputy problem and generate a markdown report for sharing with your security team.
This post looks at how Palo Alto Networks has driven convergence between SOC and cloud security. Through a series of acquisitions, PANW developed Cortex Cloud, which unifies XDR, XSIAM, CSPM, CNAPP, SOAR, and AppSec, to enable security analysts to triage alerts from code to runtime while reducing response times and costs. Cortex Cloud will still need to face challenges in onboarding automation playbooks and resolving differing workflows between SOC teams, devs, and cloud engineers.
Trusted Identity Propagation (TIP) streamlines the exchange of IdP tokens for IAM Identity Center, enabling precise access controls to AWS resources based on IdP usage and group memberships. This new SDK plugin automates the token exchange process and creation of identity-aware credentials. The plugin is currently available for the Java 2.0 and JavaScript v3 SDKs.
The BSides Seattle 2025 cybersecurity conference focused on rebuilding trust in a landscape dominated by non-human identities (NHIs). Key figures, including Wendy Nather, Heather Flanagan, and Maya Kaczorowski, emphasized that most security systems fail due to unrealistic expectations of human behavior, overlooking the fact that non-human identities, such as APIs and service accounts, outnumber humans and possess greater privileges. The conference advocated for a shift to contextual identity models based on attestation, recommending ephemeral, task-bound identities with minimal privileges that can be cryptographically verified.
Offensive Security has advised Kali Linux users to manually install a new repository signing key due to the loss of the old key to avoid update failures.
IBM announced a $150 billion investment in American innovation and advanced computing over the next five years, with a focus on mainframe and quantum computing to enhance national security and economic competitiveness.
A ransomware attack linked to the hacking group "Scattered Spider” has caused outages at Marks & Spencer.
With SentinelOne, AI isn't just a buzzword—it's the unifying force and underlying strategy that gives you holistic autonomous protection.
Sponsored
SentinelOne is introducing a complete CNAPP, powered by Generative
Multiple Critical Vulnerabilities have been identified in Planet Technology's industrial switches. CVE-2025-46275 and CVE-2025-46271 allow complete control of the router, while CVE-2025-46273 reveals hard-coded communication credentials between the NMS and managed devices, enabling remote interception and configuration changes. There are also vulnerabilities affecting specific switches. Researchers also found lax security practices, including the use of default or weak credentials for MQTT and MongoDB.
SAP Netweaver software has an unrestricted file upload vulnerability identified by CVE-2025-31324. After querying internet server search engines Shodan and Censys, Onapsis estimates that about 10,000 SAP instances are potentially vulnerable. It is a CVSS 10 vulnerability, remotely exploitable through HTTP unauthenticated, that allows for complete system compromise.
A new technique called "Policy Puppetry” can bypass safety guardrails in all major generative AI models. The attack involves crafting prompts that appear as policy files, tricking AI systems into interpreting them as instructions that override safety alignments. This universal vulnerability demonstrates fundamental flaws in AI model training and security.