Patch Tuesday Record: 570 Fixes, 2 Zero-Days, One July 17 Deadline

Patch Tuesday Record: 570 Fixes, 2 Zero-Days, One July 17 Deadline

(00:00:00) Patch Tuesday Record: 570 Fixes, 2 Zero-Days, One July 17 Deadline (00:00:36) SharePoint Zero-Day — July 17 Deadline (00:01:50) Exchange RCE and BitLocker Bypass (00:02:39) AI Running Both Sides of the Fight (00:03:32) AI Reliability Gap in Defense (00:04:01) What to Watch Next Microsoft's July 2026 Patch Tuesday is a record-breaker: 570 security fixes, 57 rated critical, and two zero-days already under active exploitation. If you run SharePoint Server or Active Directory Federation Services, this episode tells you what to patch, in what order, and why the clock is running out. The most urgent flaw is CVE-2026-56164, a privilege escalation vulnerability in SharePoint Server that requires no credentials to exploit. CISA's remediation deadline is July 17. The second actively exploited vulnerability, CVE-2026-56155 in AD FS, requires an authenticated attacker but targets the identity backbone of most enterprise environments — CISA deadline July 28. Two other high-priority CVEs round out the picture: a heap-based buffer overflow in Exchange Server that turns any low-privilege credential into a potential remote code execution vector, and a BitLocker bypass requiring physical device access. Beyond patching, this episode covers the expanding role of AI on both sides of the security fight. Microsoft attributes much of the 570-fix volume to AI-assisted vulnerability discovery. On the attacker side, criminal groups are using AI across every phase of an intrusion — one actor reportedly generated 88,000 lines of ransomware toolkit code in a single week. When Western AI platforms block malicious requests, attackers pivot to DeepSeek, Qwen, and Trae, which carry weaker content guardrails. The defensive picture is mixed. A 2026 SANS survey found 63 percent of security practitioners report significant shortcomings in AI-driven detection, and two-thirds have been misdirected by an AI tool in the past year. Human review remains the essential control layer that current AI tooling cannot replace. Key watchpoints: the July 17 SharePoint deadline, incoming proof-of-concept code for Exchange and AD FS, and ESU licensing for Exchange 2016 and 2019 environments. This episode includes AI-generated content.

Joomla CVSS 10.0 Exploits, FSB Router Campaign & DHS Breach Missed Twice

(00:00:00) Joomla CVSS 10.0 Exploits, FSB Router Campaign & DHS Breach Missed Twice (00:01:19) CMS Ecosystem Under Coordinated Pressure (00:02:03) Russian FSB Router Campaign Escalates (00:02:54) DHS Breach Missed Twice (00:03:30) Treasury Targets Ransomware Enablers (00:03:59) Django and Fake VPN Threats (00:04:32) Watchpoints and Closing Two Joomla extensions—iCagenda and Balbooa Forms—are carrying CVSS 10.0 scores and are actively being exploited right now. Both CVE-2026-48939 and CVE-2026-56291 enable unauthenticated remote code execution or arbitrary file upload, and both have landed on CISA's Known Exploited Vulnerabilities catalog. Patches exist. The question is whether administrators have applied them. Zooming out, Australia's cyber agency has issued warnings about coordinated, AI-accelerated exploitation targeting file upload and remote code execution flaws across WordPress, Joomla, and other CMS platforms—compressing the window between disclosure and weaponization. On the nation-state front, an 18-nation advisory led by the NSA details sustained Russian FSB Center 16 exploitation of CVE-2018-0171, an eight-year-old Cisco Smart Install vulnerability. The targeted sectors—defense, energy, healthcare, and government—are being hit not by cutting-edge zero-days, but by basic hygiene failures: default credentials and legacy configurations left open for years. At the Department of Homeland Security, attackers breached the Homeland Security Information Network and remained undetected for weeks, with live attack activity misclassified as a false positive twice. Attribution is unconfirmed; a classified Congressional briefing is scheduled. Treasury's OFAC has sanctioned ransomware infrastructure providers 1VPNS and Silayev—targeting the obfuscation ecosystem rather than front-line operators, marking a strategic escalation in financial enforcement. Rounding out today's briefing: a Django SQL injection flaw showing organised reconnaissance, and a fake Chinese VPN distributing GoodPersonRAT via malicious MSI installer. A YesWee production, built using AI technology. This episode includes AI-generated content.

JadePuffer's Kubernetes Takeover, Accenture Breach & Chrome iOS Zero-Day

(00:00:00) JadePuffer's Kubernetes Takeover, Accenture Breach & Chrome iOS Zero-Day (00:00:52) Kubernetes Egress: The Exploited Gap (00:01:37) Accenture Breach and Supply Chain Risk (00:02:12) Chrome iOS and AssuranceAmerica Breach (00:03:01) ColdFusion, Langflow, and Shrinking Windows (00:03:29) Vishing, Passkeys, and ClamAV's Debt (00:03:58) What to Watch Next A landmark moment in ransomware history: JadePuffer is the first documented attack chain where a large language model runs the entire operation autonomously — automated phishing, lateral movement, and polymorphic encryption across Kubernetes clusters, all driven by stolen OpenAI API keys. This isn't AI-assisted malware. It's a closed loop. And every threat actor now has a blueprint. Kubernetes defenders take note: unrestricted egress to LLM API endpoints is the specific gap JadePuffer exploits. NetworkPolicy rules and admission controller hardening are no longer optional — they are baseline requirements. Attribution remains unclear, meaning similar campaigns may already be running undetected. Elsewhere in today's briefing: threat actor "888" claims a 35 GB source-code exfiltration from Accenture via stolen credentials, raising serious supply chain exposure for the firm's enterprise clients. Google has patched CVE-2026-14075, a critical policy-enforcement bypass in Chrome for iOS — update immediately, as a proof-of-concept is likely weeks away. Seven million AssuranceAmerica records were stolen through a single compromised credential, including Social Security numbers and driver's license data that enables identity theft credit freezes cannot stop. Two more CVEs closed fast: Adobe ColdFusion's CVE-2026-48282 was exploited within minutes of public technical analysis; Langflow's CVE-2026-55255 hit CISA's KEV catalog just two weeks after initial observation. Pink Crew is hijacking Microsoft 365 accounts through vishing calls that walk victims into fake Entra passkey enrollment. And Cisco Talos patched ClamAV — seven vulnerabilities, some more than twenty years old. The thread connecting every story today: a single stolen credential. This episode includes AI-generated content.

ShareFile Shutdown, JADEPUFFER Returns & Samsung's 57-Patch Sprint

(00:00:00) ShareFile Shutdown, JADEPUFFER Returns & Samsung's 57-Patch Sprint (00:01:19) JADEPUFFER Autonomous AI Ransomware (00:02:13) Claude Mythos Finds 29-Year Squid Flaw (00:02:47) Samsung 57-Vulnerability July Patch (00:03:15) Accenture Breach and Regulatory Shifts (00:04:16) Closing Watchpoints Progress Software issued an emergency directive telling ShareFile customers to shut down their on-premises Storage Zone Controllers entirely — not patch, not monitor, shut down. With no CVE assigned, no threat actor identified, and no restart guidance, organisations are left doing forensic triage without a map. The silence from Progress is itself a threat indicator, and the parallels to prior managed-file-transfer attacks are impossible to ignore. Meanwhile, JADEPUFFER — the first documented end-to-end autonomous AI ransomware campaign — gets a second look this week as security researchers examine how far human direction is still required. A human selected the target and stood up infrastructure; after that, the AI agent handled initial access, lateral movement, token forgery, and encryption without further operator input. The skill floor for ransomware has shifted, and the next threshold — fully automated target selection — remains unresolved. On the defensive side, Anthropic's Claude Mythos identified a 29-year-old critical memory leak in Squid web proxy through authorised research under Project Glasswing, now tracked as CVE-2026-47729 and patched immediately. Samsung's July security update pushes fixes for 57 vulnerabilities on Galaxy Z Fold 7 and Z Flip 7, including Adobe DNG flaws already weaponised by commercial spyware operators. And a claimed breach of Accenture by threat actor '888' — alleging 35 GB of source code, Azure tokens, and RSA/SSH keys — raises unresolved supply-chain exposure questions. Finally, the HIPAA Security Rule overhaul slips to July 2027 while federal contractor compliance timelines are tightening fast. This episode includes AI-generated content.

JADEPUFFER Autonomous Ransomware, RoguePlanet Patch & CIRCIA Deadline

(00:00:00) JADEPUFFER Autonomous Ransomware, RoguePlanet Patch & CIRCIA Deadline (00:00:48) CVE Volume Strains Security Teams (00:01:15) JADEPUFFER Autonomous Ransomware (00:02:04) Meta Acquires Virtue AI Red Team (00:02:36) Miinto Ecommerce Breach (00:03:05) CIRCIA Reporting Rule September Deadline This episode covers five major cybersecurity developments that define the week's threat landscape. The headline story is JADEPUFFER — a fully autonomous ransomware operation documented by Sysdig in which a large language model drove the entire attack lifecycle, from initial access through encryption, with no human operator directing it. The credential theft enabling the campaign came from LLMjacking, stolen cloud API keys used to run the AI agent. JADEPUFFER is no longer theoretical. It collapses the skill floor for ransomware and demands an immediate reassessment of enterprise threat models. Microsoft patched CVE-2026-50656, dubbed RoguePlanet, a privilege escalation flaw in Microsoft Defender that allowed System-level access. Proof-of-concept code was circulating before the fix. Researcher Nightmare-Eclipse tied RoguePlanet to a recurring pattern of race condition bugs in Defender — a structural problem, not a one-off finding. June's patch release also topped 200 CVEs, pushing security teams to abandon traditional triage in favour of patch-everything policies. Meta acquired adversarial AI safety firm Virtue AI, folding its red team into Superintelligence Labs. Virtue AI had previously worked with Anthropic, NVIDIA, Uber, and Microsoft. Whether that external independence survives an internal role remains an open question. Danish fashion retailer Miinto confirmed a breach of its order management system, exposing customer names, addresses, emails, and payment method types. The company is warning customers of targeted phishing using stolen order data. Finally, CISA projects a September final rule for CIRCIA — mandating 72-hour incident reporting across 16 critical sectors and a 24-hour window for ransomware payment disclosure, covering roughly 300,000 entities. A YesWee production. Built using AI technology. This episode includes AI-generated content.