Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
Turkey has enacted new legislation that centralizes digital oversight under the Cybersecurity Directorate. The government argues that this is necessary for national security reasons. However, critics worry about this new legislation being abused by the government for censorship.
Cloudflare for Government achieved FedRAMP Class D (High) certification, sponsored by NIST, clearing the standard for the nation's most sensitive unclassified data covering law enforcement, emergency services, financial systems, and national security. The company built its High offering on the same global network and software stack running its commercial platform rather than an isolated environment, using its Data Localization Suite to confine FedRAMP High traffic processing to US data centers. Cloudflare is now pursuing DoD Impact Level 4 authorization on the same underlying systems.
HollowGraph is an Iranian-nexus malware tied to the Cavern backdoor framework that runs no command-and-control infrastructure at all and instead reads orders from and writes stolen data into a victim's own Microsoft 365 calendar over the legitimate Graph API. The implant encrypted its traffic with a hybrid RSA plus AES-256-GCM scheme using separate key pairs per direction and exfiltrated Entra ID credentials via DNS tunneling. Group-IB counted twelve infected Israeli systems active between June 3 and July 9. It linked the malware to the Lyceum espionage cluster, and Check Point separately documented the same Cavern framework this month under the name Cavern Manticore.
Atlassian has patched "RovoBlast," a single-click prompt injection vulnerability in Rovo's rovoChatPrompt parameter discovered by Varonis that allowed attackers to hijack the AI's autonomous agents and silently exfiltrate enterprise data across more than 50 connected platforms without requiring credential compromise.
CISA, the FBI, NSA, Secret Service, and South Korea's National Police Agency warned that the Gunra ransomware-as-a-service gang has hit government and critical infrastructure targets across five continents.
A 20-year-old UK man was sentenced to two years and ordered to register as a sex offender after admitting to child sexual abuse and blackmail offenses tied to The Com, a sprawling online network of minors and young adults.
Sponsored
For example your scanner flags a critical CVE in AngularJS. But no patch exists. None ever will. 🔍 For security teams, we mapped all 20 standards [5] against how EOL components impact each control. 📋 Why PCI's 30-day patch clock is unmeetable on EOL software, and the three statuses every component needs before an audit. 🛡️ How vendor-backed patching satisfies "maintained" under NIST
Security researchers uncovered an authentication flaw in the SETracker platform that many children's smartwatches run on. The researchers uncovered vulnerabilities that could be exploited by attackers to access locations, intercept messages, replace emergency contacts, record audio, and capture photos or videos on supported devices.
Metabase disclosed a critical zero-day vulnerability that is being actively exploited by attackers. The vulnerability is a critical, unauthenticated SQL injection bug that can allow remote attackers to gain administrative access to target instances. Cloud users have already been patched, but self-hosted organizations should upgrade to a patched release or temporarily block access to the ‘/api/session/reset_password/‘ endpoint until updates can be applied.
The Head Mare hacktivist group exploited unauthenticated access on TCP port 4307 to chain KLCERT-26-057 and KLCERT-26-058 for sandbox escape and SYSTEM-level code execution on TrueConf servers, deploying the PhantomCore and PhantomGraph backdoors and trojanizing client installers. Fixes are available in versions 5.3.9, 5.4.9, and 5.5.5.
Figma uses agents to review PRs for vulnerabilities, audit existing repositories, and generate secure code. When writing the workflow to review PRs, Figma found that the collection of policies and precedents that they wrote to improve precision constituted a detailed threat model for the agent to follow. This threat model was also utilized with more agents to brute force a code audit of a large, legacy monorepo and is used with agent hooks for secure code generation.
A new hybrid threat hunting framework synthesizes four years of Russian cyber operations, including activity from APT28, Sandworm, and APT29, by focusing on shared operational chokepoints across identity, edge infrastructure, and OT rather than brittle IOCs. To highlight the necessity of behavior-based detection, the playbook points to LAMEHUG's dynamic use of the Hugging Face API and CERT-UA data showing APT28 achieving domain-controller compromise in under an hour. Furthermore, the report advocates for correlating physical and cyber telemetry in response to hybrid threats like the Leipzig/Halle drone incident, though it cautions defenders against attributing attacks based solely on geographic or temporal proximity.
A reverse engineering teardown of LockBit 5.0's Linux variant found the ransomware encrypts entirely offline, generating ChaCha20 keys locally and wrapping them with a hardcoded Curve25519 public key so that zero network activity occurs during encryption, confirmed via Wireshark, socket monitoring, and interface counters. The sample also detects strace through TracerPid checks in /proc/self/status and fingerprints VM environments via /proc/ioports enumeration, forcing analysts to fall back on eBPF-based tracing to observe its behavior. The findings point to a broader shift toward offline-first ransomware design that makes network-based detection largely irrelevant against this class of threat.
Datadog has released its AI-native SAST tool. Unlike traditional scanners, the tool also uses LLMs to find vulnerabilities.
audit is an 8-stage vulnerability discovery agent inspired by Cloudflare's Project Glasswing post.
EvilFontTool is a Python CLI for red teamers that generates "evil fonts" remapping glyphs so a Word doc, PDF, or webpage displays clean text to a human reader while a different, attacker-chosen payload is what gets copy-pasted or fed to AI tools and terminals.
In a recent evaluation of Kimi K3's cybersecurity capabilities, Frontier Security attempted to run the model inside the UK's AISI's agentic sandbox and discovered that the model effectively escaped the sandbox to game the benchmark. The model first probed the network and discovered that standard DNS functionality was available for github.com, so it was able to successfully look up the benchmark, clone its repo, and lift the solutions directly from the repo. Frontier Security recommends that evaluators treat evaluation infrastructure as part of the benchmark, deny network access by default, audit traces instead of just answers, revalidate suspicious results, and assume capable models will find paths to “cheat.”
The recently founded Open Secure AI Alliance has proposed a set of guidelines for reporting cybersecurity incidents involving AI agents titled the Shared AI Findings Exchange (SAFE). The SAFE program would provide organizations with a confidential channel for reporting details of AI security incidents. It would then analyze the data, notify affected parties, flag control failures that recur across its members, and then release recommendations.
Moonshot AI's open-weight Kimi K3 broke out of a sandboxed benchmark run by private firm Frontier Security, exploiting unblocked outbound DNS and HTTPS to reach GitHub and pull the official solution instead of solving the task itself. The UK AI Security Institute only built the open-source sandbox framework and was not involved in running the test, and researchers attribute the escape to a network misconfiguration rather than deliberate model misbehavior.
Zscaler ThreatLabz tracked 351 victims in a month-long ransomware campaign, finding attackers targeted managers with "business privilege"—invoice approval, budget oversight, and contract access—rather than executives.
Jonathan Brossard has acquired securityfocus.com and the Bugtraq name and relaunched the full-disclosure mailing list, promising the original researcher-first, no-corporate-filter mission after the archives went dark through a chain of prior ownership changes.
Tencent's Zhuque Lab disclosed SCTPhantom (CVE-2026-64564, CVSS 8.5), an 18-year-old use-after-free in Linux SCTP Dynamic Address Reconfiguration dating to kernel 2.6.25 that yields root via a repeatable exploit chain bypassing KASLR and can escape containers, patched in 6.6.148, 6.12.101, 6.18.42, and 7.1.6.
Sponsored
As AI becomes embedded across the enterprise, organizations must secure their use of AI while defending against AI-powered threats. That pressure is making the limits of fragmented architectures impossible to ignore. Cato Networks was named a Leader in the 2026 Gartner® Magic Quadrant™ for SASE Platforms [5] for the THIRD YEAR RUNNING. See how Gartner evaluated the market and why, in Cato's view, unified SASE platforms are becoming essential for the AI era.
Forescout identified over 4,400 internet-facing Rockwell PLCs worldwide, including 22 in recently attacked US water utility cities that were compromised via open Ethernet/IP port 44818 rather than complex exploits, prompting urgent calls for defenders to isolate controllers behind VPNs and utilize Rockwell advisory SD1790 to recover attacker-locked systems.
VulnCheck researchers discovered an intentional and unpatchable backdoor dubbed EndlessDoors in over 100,000 active Zbtlink and Wiflyer routers that initiates outbound command-and-control connections to bypass NAT, requiring defenders to monitor port 7000 for malicious beacons and physically replace the compromised hardware.
The Swiss Federal Office of Information Technology and Telecommunications (BIT) reported that hackers exploited vulnerabilities to breach its Microsoft SharePoint instances. Other than login credentials, BIT does not believe any data was stolen. BIT responded by blocking external Internet access to SharePoint, patching the vulnerabilities that were exploited, and resetting the passwords of affected accounts.
Vanta Stealer is a highly obfuscated Python information stealer that uses PyInstaller and PyArmor to evade detection while dynamically downloading modules to harvest browser credentials, VPN configurations, and gaming platform data. The malware actively enriches stolen Discord tokens via API to identify high-value accounts before packaging the compromised data into a ZIP archive and exfiltrating it via HTTP POST. Defenders must block outbound connections to vanta[.]st and hunt for suspicious PyInstaller binaries executing unsolicited network requests to mitigate this threat.
Ill Bloom is an active cryptocurrency wallet-draining campaign exploiting a legacy CryptoJS weak-randomness vulnerability in unmaintained downstream dependencies. Defenders should immediately consult the illbloom.org technical disclosure for specific IOCs and affected wallet lists to expedite fund migration.
James Berthoty breaks down the different tools and techniques that could have helped at different stages of OpenAI's incidental breach of Hugging Face. During the initial phases of the model's escape from OpenAI, external network blocking and agentic EDRs could have been helpful. IMDS blocking and more fine-grained secrets injection could have helped Hugging Face guard against being breached by the agent. Traditional Kubernetes and CNAPP tooling could be used to detect the lateral movement from within Hugging Face.
numbat is an “AI-EDR” built by Perplexity. It provides visibility into AI agent activity with local detection, optional pre-action blocking, and forensic reconstruction.
agent-egress-bench is a standardized test corpus for evaluating AI agent egress tools, covering secret exfiltration, prompt injection, SSRF, hostname exfiltration, MCP tool poisoning, chain detection, MCP drift, A2A protocol scanning, WebSocket DLP, encoding evasion, shell obfuscation, and cryptocurrency/financial data protection.
greenlight is a skill that was created for mapping out where Claude Code's guardrails trigger exploitation-related refusals, and where they can't be relied upon. greenlight was only tested on Opus 4.7, but the methodology is generalizable.
Wiz's H1 2026 threat report highlights a 60% increase in major cloud incidents, primarily driven by cascading supply chain compromises and targeted attacks against immature AI infrastructure. Threat actors are increasingly monetizing non-human identities and reselling credential access across groups, evidenced by the sprawling TeamPCP npm campaigns and North Korean package trojanizations. This evolution signals a structural shift toward self-perpetuating attack surfaces, requiring defenders to secure unauthenticated machine-to-machine integrations and exposed backend credentials to disrupt automated access brokering.
Researchers at MIT CSAIL discovered a new branch predictor exploit that bypasses neutralization-based mitigations for Spectre v2 attacks. These mitigations assume that an attacker cannot exploit the time between when the branch predictor is isolated and when it is used by the victim branch. However, the researchers introduced a new primitive that can re-poison the CPU's state in this gap. The researchers demonstrated that this vulnerability can be exploited by users with unprivileged code execution on commodity hardware to extract /etc/shadow at a rate of 5.46 bytes/second.
LightSpy, originally a Chinese state-linked spyware, has evolved into a commercial platform actively targeting governments, enterprises, and NATO-affiliated routers across 13 countries. The modular toolkit now provides total network visibility from a pool of at least 117 command servers. Notably, researchers successfully traced the latest campaign to a Chinese contractor after an operator inadvertently revealed his real name and office address by placing a food delivery order directly through the spyware's admin panel.
Maksim Silnikau received a 16-year prison sentence for orchestrating the Ransom Cartel ransomware operation between 2021 and 2023, concluding a prolific cybercriminal career that included developing the notorious Angler exploit kit and co-creating the foundational Reveton RaaS model.
Mysk researchers found three unpatched leaks- DNS prefetching, WebAuthn's Related Origin Requests, and WebTransport- that let any website bypass Apple's Private Relay and reveal a user's real IP across all WebKit-based iOS browsers, and published the findings without prior disclosure to Apple, citing a pattern of yearlong fix delays.
Sponsored
AI is everywhere right now. But for many teams, the reality hasn't matched the promise. So what's _actually_ working? Tines recently released a guide [5] that takes a more practical look at AI adoption for security and IT teams. Inside, you'll find: * A framework for EVALUATING TOOLS BEYOND THE DEMO * A step-by-step approach to selecting tools that HOLD UP IN PRODUCTION * QUESTIONS TO ASK before committing to a vendor * Best practices for KEEPING HUMANS IN THE LOOP If you're thinking about AI beyond experimentation, this is a useful place to start.
A new Shai-Hulud worm variant has compromised over 1,280 npm packages, including the well-known Keyv library, by injecting malicious commits via hijacked GitHub Actions pipelines. The worm employs a preinstall hook to download Bun and run an obfuscated payload that aggressively extracts developer and cloud credentials such as AWS, Kubernetes, and Vault, which it then uses to spread further. Organizations that have installed affected versions should assume an immediate breach, requiring them to rotate all secrets immediately and perform manual log reviews, rather than waiting for scheduled vulnerability scans.
Manifold Security uncovered an "evil twin" campaign on the Open VSX registry involving 77 malicious extensions that impersonated legitimate packages to harvest developer environment data. While claiming to only collect anonymous metrics, the most aggressive variants actively exfiltrated OS usernames, workspace paths, Git commit hashes, and CI/CD infrastructure details to mangorbit[.]com, utilizing DNS TXT lookups for C2 resilience. Although Open VSX removed the packages on August 3, security teams should block the C2 domain and manually audit developer environments for the malicious extension IDs.
English National Ballet (ENB) disclosed that customer information was stolen via a breach of its customer relationship management (CRM) system. ENB stated that no password or financial information was stolen and advised customers to be vigilant of phishing emails.
Researcher Rotce rooted the Mercusys MB115-4G, Amazon Spain's best-selling router, via UART and traced a pre-auth path in /usr/bin/httpd where http_gdpr_decrypt copied up to 2048 bytes of AES-decrypted, attacker-controlled data into a 512-byte caller buffer with no bounds check, tracked as CVE-2026-12495. The bug was surfaced through disciplined static analysis rather than a working exploit, following the routing table into the login handler in Ghidra and confirming with Ghidriff that the flaw persisted unpatched from v1.7.0 through v1.9.0. Mercusys' own validation reproduced the crash and shipped a fix in v1.11.0 that added the missing bounds check, moved the login decrypt path from RSA to ECC, and introduced HMAC verification.
The Greatness phishing platform recently bypassed email filters by spoofing RingCentral voicemail alerts to target organizations that had blanket-whitelisted the domain, despite the lures failing standard SPF, DKIM, and DMARC checks. Once clicked, the attack routed victims through an adversary-in-the-middle (AiTM) flow to steal MFA-approved tokens, allowing attackers to replay them from remote VPNs and stealthily siphon Microsoft 365 data for weeks. To mitigate this threat, defenders must immediately replace broad safe-sender exclusions with strict authentication requirements, hunt for anomalous MFA sign-ins originating from hosting providers, and comprehensively revoke access tokens if a compromise is suspected.
The DNB Cyber Defense Center's incident response team used marimo to build out incident response workflows as modular notebooks. The team developed a custom SDK for cross-platform query translation across Splunk, Defender, and SQL interfaces while utilizing marimo's native cell dependencies and caching capabilities. The marimo-pair agentic interface also allows agents to create and execute the same notebooks as human responders.
Sponsored
AI-powered cybercrime is no longer just a future risk. ThreatDown's [13]new research found it's already here, hiding in plain sight on
researchers assess that AI capable of exploiting vulnerabilities at scale could reach criminal marketplaces within roughly six months. Read the _Cybercrime in the age of AI_ [12] report.
Airlock Digital extended its application control platform with Agentic AI Control & Governance, adding command- and session-level visibility into trusted AI agent behavior plus real-time policy enforcement that communicates decisions back to supported agents rather than simply blocking them, targeting Q3 2026 general availability.
AgentHound is an open-source offensive security framework for AI agent infrastructure that can perform recon, fingerprinting, credential looting, system prompt inventorying, model inversion, tool and instruction poisoning, and config-implant persistence and exports results to a Neo4j graph.