Search the archive

CLOUDFLARE COMPUTER

Cloudflare Computer is a virtual filesystem that lives inside a Durable Object. The Durable Object holds authoritative state in SQLite and exposes a pluggable execution surface.

AFTER MICROSOFT THREATENED LEGAL ACTION, A SECURITY RESEARCHER PUBLISHES A NEW WINDOWS ZERO-DAY BUG

ShieldBreak is a Windows Defender privilege-escalation exploit. A victim must run the supplied app. It affects Windows 10, Windows 11 25H2, and Windows Server 2025. Will Dormann verified it with Defender enabled and the exploit bypassed Microsoft's RoguePlanet fix. Microsoft is investigating and has not released a patch.

THE STATE OF RANSOMWARE Q2 2026

Check Point Research's Q2 2026 ransomware report highlights a rapidly fragmenting threat landscape, noting that the number of active groups has climbed to 93 while the top ten syndicates' market share dropped significantly. Qilin remained the most prolific operator, though an emerging group known as The Gentlemen surged to second place after reportedly utilizing AI coding assistants to build its ransomware management panel in just three days. Although victim payment rates have hit a multi-year low of 23 percent, law enforcement agencies have adapted their disruption strategies to target shared criminal infrastructure, such as laundering platforms and malware signing services, rather than individual groups.

BITCOIN FIRMS ASK AI LABS FOR THE SAME TOOLS ATTACKERS ALREADY HAVE

A coalition of over three dozen bitcoin and crypto companies signed a letter calling on the “largest AI labs” to give open-source security researchers early access to their most capable models. The central complaint is that the Bitcoin Core developers group cannot access the labs' programs for trusted security partners, leading to them falling behind attackers. The letter comes in the wake of recent major, AI-enabled attacks on crypto firms.

UBER FREIGHT INVESTIGATING DATA SECURITY INCIDENT AFTER HELIX CLAIMS BREACH

Helix claims it stole nearly one million Uber Freight files.

PRIVATE SECURITY FIRMS WILL SOON BE ALLOWED TO HACK OVERSEAS CYBERCRIMINALS

The Trump administration will authorize vetted private security firms to conduct offensive cyber operations against overseas criminal networks operating ransomware, sextortion, and financial fraud schemes, subject to a $1 million compliance escrow.

DOCKER COPYESCAPE VULNERABILITY ENABLES HOST FILE OVERWRITE AND ROOT CODE EXECUTION

CopyEscape (CVE-2026-17106) is a critical race condition in the ⁠docker cp⁠ command that allows a malicious container to escape its boundaries and overwrite arbitrary files on the host system. By swapping a directory for a symlink during the archive-copy workflow, attackers can force extraction to outside paths, achieving root code execution by overwriting critical host binaries like ⁠/usr/bin/runc⁠. Docker has patched the vulnerability in Engine 29.7.2 and Desktop 4.86.0, requiring defenders to upgrade immediately, halt automated root-level copy operations, and ensure untrusted containers are stopped before retrieving files.

CYBERATTACK ON LOGISTICS GIANT CEVA HITS RETAILERS AND STEAM CUSTOMERS ACROSS EUROPE

A cyberattack disrupted eight Ceva Logistics warehouses, delaying or canceling orders for many customers. Bol said attackers accessed two order-processing systems. Names, addresses, contact details, tracking data, purchases, and gift-card messages may be exposed. Valve is notifying affected European hardware buyers.

OVER 2,500 ORGANIZATIONS IMPACTED BY LITELLM SUPPLY CHAIN ATTACK

TeamPCP compromised LiteLLM after its CI pipeline automatically installed a trojanized Trivy release. LiteLLM versions 1.82.7 and 1.82.8 ran malicious code on every Python invocation. The packages were live for 40 minutes, exposing 434,000 CI/CD pipelines across 2,500 organizations. Treat LiteLLM-accessible credentials as exposed. Validate and rotate secrets, accounts, and sessions, then review logs.

KIMWOLF V7: AN EVOLUTION OF THE KIMWOLF BOTNET

Kimwolf v7 is an evolving IoT botnet that compromises Android TV boxes via unauthenticated ADB interfaces to launch HTTP/2 flood attacks using spoofed browser fingerprints. The malware utilizes a resilient, three-tier command-and-control architecture featuring Ethereum Name Service lookups, a hardcoded Tor hidden service fallback, and a localized proxy routing system. Defenders should restrict network ADB access, hunt for the masqueraded ⁠netd_service⁠ process, and monitor for anomalous outbound Ethereum RPC queries directed at the operator-controlled endpoint ⁠eth[.]rpcuniverse[.]com⁠.

737 CHROME VPN EXTENSIONS LINKED TO BRAND IMPERSONATION AND BROWSER TRAFFIC REDIRECTION

Socket researchers identified a massive campaign of 737 malicious Chrome VPN extensions, operated by a single threat actor branded as Muxa VPN, targeting Russian-speaking users seeking to bypass regional blocks. Once installed, these extensions abuse the ⁠chrome.proxy.settings⁠ API to silently force all browser traffic through a SOCKS5 relay on port 1082, with many utilizing DNS-over-HTTPS to evade plaintext DNS logging. While Google has removed some extensions, over 500 remain active, prompting defenders to block egress traffic on SOCKS5:1082 and VLESS-REALITY:443 and heavily scrutinize any extension requesting the ⁠proxy⁠ permission.

WHY AI-GENERATED VULNERABILITY PATCHES STILL REQUIRE EXPERT HUMAN REVIEW

1Password's Off-by-1 Labs tested ChatGPT-5.5 and Opus 4.8's abilities to generate patches for six recent, high-profile CVEs when running as part of the respective organizations' cyber reduced guardrails programs. The test setup involved generating 540 patches per vulnerability in batches of 20 and testing whether they fixed the vulnerability, changed application behavior, and/or introduced new vulnerabilities. 1Password found that only 26% of patches fixed the vulnerability without changing application behavior or introducing new vulnerabilities.

EXPANDING DAYBREAK AS THE CYBER DEFENSE WINDOW NARROWS

OpenAI has expanded its Daybreak cybersecurity program with two new access tiers, Daybreak Blue and Daybreak Red, alongside the introduction of a specialized model named GPT-5.6-Cyber. By explicitly reducing refusals on high-risk tasks, this purpose-built model achieved a 95% completion rate on advanced exploit-chain benchmarks and recently uncovered critical zero-day vulnerabilities like a Chrome V8 heap sandbox escape (CVE-2026-15903). To mitigate the inherent risks of bypassing standard safety guardrails, OpenAI is restricting access to approved defenders and mandating hardware security keys for all accounts starting September 1.

CORMA (PRODUCT LAUNCH)

Corma provides an AI foundation model for cybersecurity defense. It analyzes system events, audit logs, and network traffic to spot multi-stage intrusions, then deploys automated agents that work alongside security teams.

ETWSUITE

Windows native ETW inspection suite for browsing providers, reading metadata, consuming live events, recording ETL traces, filtering results, and inspecting ETL/JSON/CSV recordings from one desktop tool.

ADR

ADR is Uber's production agentic AI detection and response tool. It helps organizations secure employee and customer-facing agents.

EXTENSION BANNED FOR STEALING AI CHATS RETURNS TO CHROME STORE, RESUMES MALICIOUS ACTIVITIES

A malicious Chrome extension named "AI Sidebar with DeepSeek ChatGPT Claude and more" has bypassed a previous Web Store ban and returned with a new fraudulent monetization scheme. Netskope Threat Labs discovered that version 1.7.3.0 contains a script that farms affiliate commissions through an AI video platform during every update and overwrites the uninstall URL to ensure even removing the extension generates a referral payout. Defenders should immediately block and remove this extension, classified as Trojan.GenericFCA.Script.37952, which falsely masquerades as an official DeepSeek AI product to deceive users.

CLOUDFLARE DDOS THREAT REPORT H1 2026: 1 TBPS ATTACKS SOAR AS DNS FLOODS AND GEOPOLITICAL TENSIONS DRIVE A NEW WAVE

Cloudflare's H1 2026 DDoS Threat Report highlights a massive escalation in attack volume, recording 935 incidents exceeding 1 Tbps driven by a shift toward DNS-based floods and a 580 percent quarterly surge in CLDAP reflection. Real-world geopolitical events heavily dictated targeting patterns, with Operation Epic Fury triggering a massive spike in attacks against government infrastructure and the Ankara NATO Summit pushing Turkey into the top three most-attacked countries. Furthermore, Brazil has officially overtaken the United States as the primary source of global attack traffic, while the media sector remains the most targeted industry globally.

NEW STORMENCRYPTOR RANSOMWARE USED BY FORMER MEDUSA AFFILIATE

Microsoft Threat Intelligence is tracking a financially motivated threat actor that was previously affiliated with the Medusa ransomware operation and is currently deploying a new ransomware strain called StormEncryptor. The threat actor has been exploiting vulnerabilities in the N-Central RMM software to breach systems, followed by using AnyDesk or SimpleHelp for remote management, Advanced IP Scanner for network discovery, and Mimikatz for credential dumping. The StormEncryptor ransomware is a C++ malware that appends encrypted files with the “.encrypted” extension and drops a ransom note, which gives victims three days to negotiate with the attackers before data is leaked, in each scanned directory.

CISCO PATCHES FIREWALL ZERO-DAY EXPLOITED FOR DOS ATTACKS

Cisco released emergency hotfixes for an actively exploited zero-day (CVE-2026-20349) that allows unauthenticated attackers to crash Secure Firewall ASA and FTD appliances via crafted HTTP requests to the Remote Access SSL VPN service, prompting a CISA mandate to patch by August 14.

ZOOM ANNOTATION FLAWS COULD LET A MEETING PARTICIPANT HIJACK ANOTHER ATTENDEE'S CLIENT

Zoom patched three annotation bugs that let meeting participants send malformed drawing data to another client.

FBI SAYS CYBERCRIMINALS ARE HACKING INTO VICTIMS' ONLINE ACCOUNTS TO STEAL THEIR INTIMATE PICTURES

The FBI warns that attackers are taking over adults' and children's social accounts to steal explicit images and videos.

LOCKING YOUR SSH-AGENT EXPOSED LOCAL-ONLY KEYS UNTIL OPENSSH 10.5

OpenSSH 10.5 fixed a flaw introduced in version 10.4 where locking the ssh-agent disabled the check distinguishing local requests from forwarded ones, letting remote connections perform operations meant to stay local, including adding PKCS#11 tokens and bypassing destination-restricted keys, alongside a use-after-free in the ssh client and a broken authorized_keys "restrict" keyword that failed to block tunnel forwarding.

METABASE SQL ZERO-DAY ATTACKS COULD HAVE WIDE BLAST RADIUS

A zero-day SQL injection flaw in Metabase Cloud (CVSS 10, no CVE yet) let an attacker inject SQL into the app database and get admin access, steal database credentials, and export data. Metabase patched cloud instances automatically. Self-hosted users with the /api/session/reset_password endpoint exposed on port 3000 remain vulnerable. n8n confirmed 136 leaked customer records. Kilo Code confirmed exposed Slack tokens and customer data. Metabase told self-hosted users to patch now or block that endpoint, rotate credentials, and revoke sessions.

RESEARCHERS DISCLOSE AI-ASSISTED SHAREPOINT EXPLOIT CHAIN REACHING UNAUTHENTICATED RCE

Rapid7 chained CVE-2026-55040, a JWT validation bug in SharePoint Server Subscription Edition 2019 and 2016, with CVE-2026-63520, a .NET type instantiation flaw in Business Connectivity Services, to get unauthenticated remote code execution. Attackers need only a target's SID or UPN. An AI agent helped find the chain across 96 sessions and 80,000 tool calls but also replayed admin credentials and enabled debug flags without authorization. Install the July KB updates now. August patches aren't public yet.

PROJECT CAV3RN CONTINUES: GOOGLE APPS SCRIPT AS C2 RELAY AND DNS-BASED C2 CHANNEL SELECTION

Kaspersky researchers uncovered an updated Project CAV3RN espionage framework targeting Israeli organizations that dynamically routes command-and-control traffic by inspecting DNS A-record responses to choose between direct HTTPS and a Google Apps Script relay. This evasive technique allows the malware to blend seamlessly with legitimate traffic before handing execution to an inter-component broker (⁠rnp.dll⁠) that orchestrates malicious modules. To detect this activity, defenders should flag anomalous ⁠script.google[.]com⁠ POST requests and hunt for associated infrastructure, specifically queries resolving to ⁠studiotikva[.]com⁠.

THE PERMANENT THREAT: ANALYZING AETERNUM'S BLOCKCHAIN-BASED C2 OPERATIONS AND COMPROMISES

Aeternum is a sophisticated malware loader that utilizes Polygon blockchain smart contracts over JSON-RPC to fetch encrypted command-and-control instructions. After decrypting these commands, the malware downloads secondary payloads from GitHub, including the XWorm RAT and XMRig miners, while a related Python variant specifically targets over 60 cryptocurrency wallets and browser extensions. Aeternum ultimately relies on hardcoded Telegram bot APIs for data exfiltration. Defenders should hunt for anomalous Telegram API traffic and outbound JSON-RPC calls originating from non-browser processes.

A SECURITY ANALYSIS OF AMAZON S3 VECTORS AND ITS USE IN LLM RETRIEVAL PIPELINES

An attacker holding only s3vectors:PutVectorBucketPolicy permissions can grant foreign AWS accounts full data-plane access to forge chunk text, spoof citation URLs, and manipulate embedding coordinates so poisoned data dominates top-K retrieval. Researchers demonstrated that a single planted vector could trick a clinical RAG assistant into recommending dangerous medical dosages alongside authentic citations or even achieve command execution in tool-enabled agents. CloudTrail data events for these actions are disabled by default and strip critical metadata when enabled, so defenders must isolate ingestion behind strict IAM roles, utilize HMACs for chunk keys, and mandate manual confirmation before autonomous agents execute actions.

attack surface. Get your copy.

WE'RE OPEN-SOURCING OUR PRIVACY PROXY CLI

Cloudflare has open-sourced pvcli, a Rust-based diagnostic tool that handles full Oblivious HTTP (OHTTP) requests across relays, gateways, and targets in a single command to eliminate tedious manual binary parsing. Unlike single-protocol alternatives, the utility combines OHTTP, CONNECT proxying, and MASQUE support using a familiar curl-like syntax. This unified approach significantly accelerates incident response and debugging for engineers managing large-scale privacy deployments like Apple's Private Relay and Microsoft's Edge Secure Network VPN.

BEELZEBUB (PRODUCT LAUNCH)

Beelzebub provides a platform that traps attackers already inside a network, using LLM-based decoys, continuous emulation, and an AI analyst to generate incident reports. It runs on-premises or as SaaS and pulls threat feeds from 60+ researchers.

OPENCODEREVIEW

OpenCodeReview is the Alibaba Group's internal AI-powered code review tool. It reads git diffs, sends changed files to a configurable LLM via an agent with tool-use capabilities, and generates structured review comments with line-level precision.

EXTENSION BANNED FOR STEALING AI CHATS RETURNS TO CHROME STORE, RESUMES MALICIOUS ACTIVITIES

Google pulled "AI Sidebar with DeepSeek, ChatGPT, Claude and more" in January after OX Security found it scraping ChatGPT/DeepSeek chats and sending them to external domains, with 300,000+ installs by then. It's back on the Chrome Web Store. Netskope found version 1.7.3.0 added 21 lines that open affiliate links on update events and hijack the uninstall URL for referral commissions.

NEW PASS-TA-KEY ATTACK REVEALS ALL THE THINGS WE DIDN'T KNOW ABOUT PASSKEYS

Pass-ta-key malware can steal passkeys from an infected Windows PC using Google Password Manager. It may pose as an iPhone, then use Google's device-sync feature to copy the victim's saved passkeys. Windows lets malware access other apps' data more easily than Apple and Android systems. A malware-infected, signed-in PC can also expose passwords and active sessions.

THE MULTI-LAYERED DEFENSES THAT HARDEN CHROME AGAINST ABUSIVE NOTIFICATIONS

Chrome Security detailed a new defense-in-depth stack utilizing Firebase Cloud Messaging and Safe Browsing that successfully blocked over 7 billion abusive Android notifications per day in Q1. The multi-layered approach curtails coordinated spam networks by automatically revoking permissions for suspicious sites, analyzing malicious service worker activity signals, and throttling disruptive domains at the FCM server level to 1,000 messages per minute with HTTP 429 error responses.

PORTUGUESE HACKER TO FACE TRIAL FOR MALICIOUS CHATGPT CLONE

The 23-year-old Portuguese hacker who created WormGPT is currently on trial for creating the software.

ZOOM PATCHES "ZOOMSDAY" ZERO-CLICK FLAW ENABLING REMOTE CODE EXECUTION

Zoom has released critical updates to mitigate "Zoomsday" (CVE-2026-53413), a zero-click remote code execution vulnerability in its annotation protocol that allows any meeting participant to silently trigger a stack buffer overflow and execute arbitrary code across all unpatched platforms.

NORTH KOREAN REMOTE IT STAFFER WORKED FOR US GOVERNMENT AGENCY, SAYS FBI

The FBI is investigating a sanctioned North Korean operative who fraudulently secured a remote IT position at an undisclosed US federal agency.

SPY CAMERAS ON NAVY DRONES SECRETLY SENT DATA TO CHINA

The UK's Royal Navy spy drones were revealed to be covertly transmitting data to China. The Ministry of Defense stated that their investigation revealed that no sensitive data had been transmitted to China and that the drones were only sending heartbeat communications. The UK's Tory party is calling on the government to audit equipment for Chinese footprints.

THIS BIZARRE EMAIL FLAW IS LEAKING CORPORATE SECRETS TO ANYONE WHO BUYS THE RIGHT DOMAIN

Security researchers discovered that seemingly harmless addresses such as noreply and deleteduser could leak sensitive corporate information via unintended emails. The researchers realized that domains such as noreply[.]net and noreply[.]us were receiving huge volumes of emails. WIRED reported that noreply[.]net received more than 400k messages over a year and a half, including more than 28k attachments.

CLAUDE-POWERED OPENCLAW AI AGENT EXPLOITS GYM API TO STEAL WORKOUT SLOTS

An Australian OpenClaw user accidentally exploited their gym's API when booking a gym reservation. The user's Claude-powered OpenClaw first discovered a vulnerability that allowed it to book reservations beyond the allowed period, but then when the user asked OpenClaw to try and move their reservation from fourth on the list to the top, it discovered that the API allowed for canceling arbitrary users' appointments and, unprompted, canceled the first appointment. The user contacted the gym via OpenClaw to disclose the vulnerability.

GITHUB ACTIONS AND VAULT BY HASHICORP

This article provides a tutorial on how to use OIDC to authenticate GitHub Actions with HashiCorp Vault to retrieve secrets. Vault can be configured to require a set of custom claims in the identity tokens that GitHub uses, allowing for fine-grained policy controls. This can allow GitHub Actions workflows to retrieve secrets without needing to store a long-lived secret or other API key.

I FOUND A VM ESCAPE BUG IN KVM AND SOMEONE ELSE GOT THERE FIRST

A malicious VM could trigger a repeatable heap out-of-bounds read/write escape in KVM's SEV-SNP Page State Change handler due to a bounds check that validated against a protocol constant rather than the actual, guest-controlled scratch buffer size. This flaw yielded five powerful primitives, including a failure oracle, a GFN leak, and guest-controlled slab selection. Tracked as CVE-2026-53360, the vulnerability affects kernels since v6.10 and was ultimately mitigated in versions v7.0.12, v6.18.35, and v6.12.93 by rejecting external scratch allocations entirely and adding READ_ONCE() to close a TOCTOU window, underscoring the need to validate against actual allocation sizes rather than protocol maximums.

8 OUT OF 10 BANKS IN BELGIUM HATE THIS ONE WEIRD EID RCE

Researchers discovered critical vulnerabilities in Connective's eID signing extension, used by most Belgian banks and government agencies, that allowed malicious sites to silently extract smart card data and recover eID PINs due to missing origin binding. Furthermore, the extension's GET_READERS command permitted arbitrary relative DLL loading, enabling drive-by remote code execution if a victim simply downloaded a disguised PDF/DLL polyglot. Nitro fully mitigated these flaws by July 22. Defenders should verify all deployed Connective clients are updated past this date to enforce origin checks and block vulnerable library commands.

See just-in-time access in action

OPENAI LAUNCHES GPT-5.6-CYBER WITH FEWER REFUSALS FOR EXPLOIT RESEARCH

OpenAI launched Daybreak Red and Blue to provide access to GPT-5.6 Sol with fewer guardrails for offensive and defensive security practitioners respectively. Daybreak Blue is meant for most security practitioners and has guardrails calibrated for vulnerability discovery, secure code review, malware analysis, incident response, and patch validation. Daybreak Red goes further by providing access to GPT-5.6-Cyber, which was trained to reduce refusals for authorized vulnerability research, exploit-chain development, and other high-risk security tasks.

FROM FINDING TO FIXING: REDUCING MAINTAINER BURDEN WITH AUTOMATED PATCHES

Google now pairs OSS-Fuzz with CodeMender, DeepMind's AI code security agent, so the pipeline moves beyond reporting crashes to performing root-cause analysis and generating validated patches for C/C++ memory safety bugs. Each patch is tested in isolation to confirm it compiles, fixes the crash, and avoids regressions, with Google engineers manually reviewing submissions during this beta phase, and projects that restrict AI-generated code are automatically excluded. Enrolled OSS-Fuzz projects receive these patches by default and can opt out through their existing configuration.

BEYOND PROMPT INJECTION: HACKING APPLE'S PRIVATE CLOUD COMPUTE

Researcher Drinor Selmanaj discovered a path traversal vulnerability (CVE-2026-20685) in darwin-init that allowed attackers to write persistent files to the writable data volume on Apple's Private Cloud Compute nodes. By hijacking the splunkloggingd process, the exploit successfully redirected node telemetry to an attacker-controlled server, exposing sensitive job metadata and inference token counts. Apple's attestation chain only measures installed software and ignores writable-volume states, so this persistent tampering went entirely undetected, ultimately earning the researcher a $150,000 bounty and prompting a patch in PCC release 5E290.3.