🔴 RingCentral breach confirmed at 1.6M accounts, 🧾 Beacon CRM donor data out via leaked AWS key, 📦 Trezor disclosure narrows ShipMonk breach window, ⚠️ GeoServer zero-day exploited before patch, 🎯 Dysphoria botnet tops 296k IoT devices, 🐍 DCRat rebuilt in-browser from SVG

The day's defining story is the RingCentral data breach: Have I Been Pwned now puts the July intrusion at 1.6 million accounts, confirming ShinyHunters' claim that it stole 623GB from the cloud communications provider and leaked a 280GB archive after RingCentral refused to pay the ransom, so businesses on its calling, messaging, and voicemail platform should treat breach-themed phishing as likely and watch for unauthorized account activity. The Beacon CRM data breach exposed the entire customer database of a platform used by more than 1,000 UK charities after a leaked AWS access key sitting in public JavaScript build artifacts gave the attacker access to data and attachments, a supply-chain class breach for the non-profit sector, so assume donor personal data is out and scan repositories and build artifacts for committed credentials.

Microsoft's August Patch Tuesday release covers roughly 400 CVEs, led by the exploited AFD.sys WinSock elevation-of-privilege flaw CVE-2026-68820 that CISA added to its Known Exploited Vulnerabilities catalog on Aug 11, and follow-on coverage this week adds the Exchange Server branch, six flaws spanning remote code execution, privilege escalation, denial of service, spoofing, and a security-feature bypass disclosed Aug 11 with CVE-2026-62913 updated the next day, plus the Windows 10 KB5120249 and Windows 11 KB5121003 cumulative updates and a CVSS 8.8 Outlook RCE in CVE-2026-70329 reachable via crafted Office files. The same release finally carries official patches for LegacyHive, the Windows User Profile Service privilege escalation disclosed in mid-July by Nightmare Eclipse: a non-admin user can modify the classes registry hive for code execution at the next administrator login on fully patched builds, so verify the fix is deployed and hunt the published Defender for Endpoint queries.

The Trezor breach at ShipMonk exposed order-related personal data for roughly 13,689 customers, 11,742 of them with full name, email, phone, and shipping address, after ShipMonk notified Trezor of unauthorized access to its systems on Aug 10; Trezor's own Aug 13 disclosure puts the affected window at orders shipped May 10 through Aug 8 across the US, UK, Sweden, Colombia, Brazil, Italy, and Portugal. Private keys, wallet backups, and holdings were never at risk, so the exposure is downstream fraud, and Trezor customers should treat unsolicited contact referencing a real order as hostile by default. Apple sent a new batch of high-confidence Threat Notifications on Aug 13 to iPhone users in 110 countries, warning of individual targeting by mercenary spyware without naming the family, so recipients should treat the alert as active-monitoring risk and consider lockdown mode.

Jamf Threat Labs uncovered AmnesiaStealer, a macOS infostealer delivered through a ClickFix campaign fronted by a counterfeit GitHub portal at github.aoitour[.]com that asks victims to paste a Terminal command and enter their password; the payload steals credentials and browser data and silently takes control of live Chromium sessions, so never paste commands into Terminal from a web page and alert on browser-initiated Terminal execution. Attackers began exploiting an unpatched GeoServer zero-day within hours of its disclosure, an SQL injection in the jsonArrayContains filter usable with PostGIS and Oracle JDBC data stores that can reach remote code execution, with no CVE assigned and no patch available, so identify and quarantine exposed instances now.

Fortinet's August patch wave fixes an unauthenticated FortiWeb login bypass (CVE-2026-26035, a CWE-287 improper-authentication flaw that lets a remote attacker log in to the GUI or CLI with a random username and password when the wildcard admin setting is enabled) and arbitrary code execution in FortiClient for Windows (CVE-2026-70465), while the CCCS advisory AV26-812 adds CVE-2026-70468, an authentication bypass that lets a remote attacker impersonate any FortiGate managed by FortiManager; no exploitation is reported yet, but disable the FortiWeb wildcard setting and patch FortiManager and FortiClient as the priority. The vCenter campaign exploiting CVE-2026-59310, a CVSS 9.8 directory traversal with no workaround, keeps expanding as more than half of the 361 compromised IPs across 47 countries sit in Germany, the US, Turkey, Iran, and France, with an assessed APT operator planting reverse_ssh backdoors and a generic YARA rule published for the tool.

President Trump signed a national security presidential memorandum that lets vetted private US companies run offensive cyber operations against transnational criminal organizations, with DHS and DOJ program executive directors granting authorizations and 60 days to draft the company-vetting standards; critics warn the arrangement legitimizes hack-back, so the rules of engagement and liability for cyber effects are the near-term question. The Fairlife ransomware attack became the case study in why patching alone is not a complete defense: the Anubis group entered the Coca-Cola-owned dairy through CVE-2025-5777, the CitrixBleed 2 flaw already on CISA's KEV, encrypted Nutanix systems, stole roughly 1 TB of data, and forced a production suspension across US facilities. Huntress detail on the Akira intrusion shows one missing MFA control escalating a credential spray into a full compromise: the affiliate entered through an MFA-less SonicWall VPN, rebooted into Safe Mode with Networking to disable EDR and Defender, archived shares with WinRAR, exfiltrated via s5cmd, and stole data even though the encryption stage failed.

A tech contractor for Brightly Software was sentenced to two years in prison after stealing payroll and compensation data as an insider and extorting the Siemens-owned SaaS company for a $2.5 million threat of which he collected a small fraction, a compact demonstration of the insider-threat failure mode of trusted data access without monitoring. Leak-site volume stayed high: Clop claimed AOL.COM on Aug 12 and threatened to publish unless negotiations start, Aurora claimed Austrian jeweler FREYWILLE with 2025 product costing and enamel color recipes among the leaked files, Majinahanashi leaked 10,382 files from French audio retailer SON-VIDEO, Emperador claimed the City Government of Baguio with official contracts and financial statements, and Qilin threatened US trade union Local 345 with a full leak. Check Point's State of Ransomware Q2 2026 shows the ecosystem fragmenting: the top 10 groups account for 57.6 percent of victims, down from 71 percent, the active group count rose to a record 93, and Qilin stayed most prolific with 279 victims.

The malware-crypting market has matured into a subscription economy: cybercriminals can now rent payloads that re-encrypt themselves to keep evading antivirus, with Cruciferra, a crypter-as-a-service platform tied to tax-themed email campaigns against Indian taxpayers, wrapping customer payloads in evasion layers, and Recorded Future's Insikt Group profile of 24 crypting vendors showing in-memory execution, anti-analysis controls, process injection, and rapid re-crypting sold as packaged services. The same commoditization logic is hitting AI credentials: poisoned npm packages harvest AI API tokens from developer build environments for resale through gray-market proxies, with at least one victim left with a seven-figure bill, so scope AI keys to minimum quota, rotate anything exposed in build logs, and alert on anomalous token spend.

Two new threat-actor developments landed today: the Dysphoria botnet has expanded into a major Internet of Things threat, with a new Shadowserver report counting roughly 296,000 compromised routers, gateways, and IP cameras running embedded Linux and repurposed for DDoS attacks plus residential proxy and relay operations, so inventory and harden exposed network gear. Trellix dissected a DCRat campaign that turns an apparently harmless SVG attachment into a full delivery chain, reconstructing a password-protected malware archive directly inside the victim's browser from a Colombian judicial-themed lure, so treat SVG attachments as executable content and block browser-initiated archive extraction.

Akamai researchers presented a Bring Your Own EDR attack at DEF CON 34 that turns a legitimate SentinelOne agent into a privileged trojan on Windows: exposed COM interfaces, specifically the SentinelHelper.1 Dump method, let a local administrator dump PPL-protected processes including MsMpEng.exe, execute unsigned code inside a PPL process, and enroll agents that look healthy but lack management connectivity, fixed in Agent 26.1.1. Researcher Christopher Domas released a DRAM scrambling attack against AMD Family 15h and 16h processors that remaps memory-controller address translation to expose protected DRAM regions, attacking the hardware isolation boundary itself in a way invisible to endpoint defenses, with firmware updates and hardware refresh the only real mitigations. Aeternum is a new botnet running its command-and-control through smart contracts on the Polygon blockchain: infected Windows hosts poll public RPC endpoints for instructions that no server seizure can erase, so detection moves to JSON-RPC traffic to blockchain endpoints and Startup persistence under misleading names.

On the espionage side, Jewelbug, a China-based hackers-for-hire group also tracked as Earth Alux, runs espionage and crypto fraud from one control panel: the Antino Windows implant, XG-Web tool, a malicious PDF Viewer browser extension, and the Rust ClientKing with 37 builds fed a victim database of more than one million implant check-ins and 580,000 stolen browser cookies in under three months, including a webmail attack that planted a malicious script in a government mail template. An APT36-linked campaign against Afghan telecom and South Asian critical infrastructure uses three undocumented families, PATCHCORD, SHEETCORD, and the HACKERAI C2 agent, delivered via a malicious Inno Setup package masquerading as Afghan Telecom software, with code style suggesting LLM-assisted development. The HoneyMyte group upgraded CoolClient, the Mustang Panda-linked backdoor, with a variant that can deploy a signed kernel-level Windows rootkit in campaigns across Asia and Russia, and Armored Likho extended its Russia-targeting toolkit with GitHub as a backup C2 channel and automated microphone eavesdropping plus Telegram session hijacking on compromised Windows endpoints.

The LiteLLM supply-chain compromise attributed to TeamPCP keeps expanding: new forensics trace the entry through a compromised GitHub Actions pipeline of Trivy that stole PyPI publishing tokens for malicious LiteLLM 1.82.7 and 1.82.8, with the reconstructed blast radius at more than 2,500 organizations and roughly 434,000 CI/CD pipelines, so rotate cloud credentials and audit pipeline secrets as if the dependency chain was compromised. A scan near more than 1,000 US data center locations found 548 internet-exposed building automation devices running end-of-life products, Tridium NiagaraAX 3.x, Trane Tracer SC, and Carrier WebCTRL 7.0, with no future security patches, a chronic physical-infrastructure risk where a compromised BAS controller can disrupt cooling or power. Ukrainian authorities shut down 94 fraudulent call centers in an operation with German police, seizing 1,794 workstations and millions in cash from fake-investment and bank-fraud schemes.

Cisco Talos identified an undocumented phishing framework internally branded JWR that impersonates checkout and login pages and keeps an AES-CTR encrypted WebSocket open so an operator steers each victim session live, harvesting identity documents, Social Security numbers, 2FA codes, and full device fingerprints, with a real-world campaign delivered by SMS lures impersonating toll authorities and couriers in Southeast Asia and the Middle East. SentinelLABS aggregated four frontier-lab agentic intrusions from OpenAI, Anthropic, and Meta, plus the UK AI Security Institute, in which AI agents sustained intrusions across days and wrote fresh disposable tools per session, arguing that a model does not choose its objective or granted access, so agent behavior and tool churn need their own telemetry; XBOW research adds that AI's mid-tier models such as GLM-5.2, Grok 4.5, Opus 4.7, and Muse Spark 1.1 have crossed the practical threshold for offensive work at a cost that makes brute-force repetition viable.

The advisory queue stayed busy: Dell patched an OS command injection in the VSI IAPI component used by the vSphere Client (CVE-2026-67261, CVSS 9.8, root-level, fixed in 10.11.1.0 via DSA-2026-335), GitLab 19.2.2, 19.1.4, and 19.0.6 close the issues behind CCCS advisory AV26-814 with no CVE identifiers named yet, WebPros Plesk has a privilege escalation via database cloning fixed in 18.0.79.6 and 18.0.80.2, and Wireshark 4.6.8 patches 28 issues in the dissector stack, so treat unsolicited capture files as untrusted input.

Watchlist: check whether RingCentral or Beacon CRM data appears in the public dumps and harden exposed routers and cameras against Dysphoria, patch the AFD.sys and Exchange branches and verify LegacyHive is deployed, quarantine exposed GeoServer and vCenter instances and patch FortiWeb, FortiManager, and FortiClient, require MFA on every VPN given the Akira and Gunra patterns, keep offline and immutable backups while leak-site volume holds, hunt blockchain RPC polling, PPL-dump, and Chromium session-control behavior, and treat AI provider keys, donor databases, and hardware-wallet order data as high-value credentials under active targeting.