Rosetta Daily · Jun 2, 2026

Auto-generated · 29 sources configured (WebSearch mode — feed allowlist still in effect) · 15 items selected
Window: past 24 hours (2026-06-01 → 2026-06-02)

In-the-Wild / Actively Exploited

• 🔴🔥⚠️ WP Maps Pro CVE-2026-8732 (CVSS 9.8) — unauth admin-account creation, exploited at scale across 15,000+ WordPress sites
  A privilege-escalation flaw in the WP Maps Pro plugin lets an unauthenticated attacker invoke the wpgmp_temp_access_support() "temporary support access" function without adequate checks and mint a full administrator account — complete site takeover. Affects all versions ≤ 6.1.0; fixed in 6.1.1. Wordfence reports active exploitation already underway: 2,858 attacks blocked in 24h, with 3,600+ exploitation attempts in a single day, and 15,000+ installs exposed. The vendor's temp-access email is reportedly hardcoded. Update to 6.1.1 immediately and audit for rogue admin users.
  The Hacker News · Bleeping Computer · TechRadar

• 🔴🔥 Microsoft Defender CVE-2026-41091 & CVE-2026-45498 — KEV remediation deadline is tomorrow (June 3)
  Two actively-exploited Microsoft Defender flaws — CVE-2026-41091 (local elevation of privilege via improper link resolution) and CVE-2026-45498 (denial of service) — sit in CISA KEV with an FCEB deadline of 2026-06-03. Fixed in Defender Antimalware Platform 1.1.26040.8 / 4.18.26040.7. Verify the platform actually auto-updated rather than assuming it did — Defender's silent-update path is the failure mode here.
  The Hacker News · Malwarebytes · CISA KEV

• 🔴🔥⚠️ PAN-OS GlobalProtect Authentication Bypass (CVE-2026-0257) — exploitation expanding to breach corporate networks
  Palo Alto Networks continues to warn that the GlobalProtect portal/gateway auth-bypass (CVSS 7.8) is being exploited to establish unauthorized VPN connections; CISA's KEV deadline for it has now passed (added 5/29, due 6/1) and exploitation is broadening. Affects firewalls with auth-override cookies enabled under specific certificate configs. Patch the fixed PAN-OS maintenance releases; disable auth-override cookies as interim mitigation.
  The Hacker News · Palo Alto advisory · CISA KEV

Critical Vulnerabilities & Advisories

• 🔴 Apache HTTP Server CVE-2026-23918 (CVSS 8.8) — HTTP/2 double-free: trivial DoS, possible unauth RCE
  A double-free in mod_http2 (stream cleanup path in h2_mplx.c) fires when a client sends an HTTP/2 HEADERS frame immediately followed by RST_STREAM with a non-zero error code before the multiplexer registers the stream — the same h2_stream pointer is pushed onto the cleanup array twice. DoS is trivial: one TCP connection, two frames, no auth, and the worker crashes on any default mod_http2 + multi-threaded MPM deployment. The RCE path requires the APR mmap allocator — the default on Debian-derived systems and the official httpd Docker image, so a large share of real deployments are RCE-exposed, not just DoS-exposed. Affects 2.4.66; patch to 2.4.67. Researchers demonstrated working PoC RCE in lab conditions.
  The Hacker News · Security Affairs · SOCRadar

• Apache Flink CVE-2026-35194 — SQL injection in code-generation engine → RCE
  A high-severity SQL-injection flaw in Apache Flink's code-generation engine exposes distributed data-processing environments to remote code execution. Internet-reachable or multi-tenant Flink clusters should be treated as crown-jewel-adjacent; patch per the Apache advisory and restrict job-submission surfaces.
  FreeBuf vuln portal

Vendor Advisories

• cPanel CVE-2026-29201 / 29202 / 29203 — critical privilege-escalation, code-execution and DoS trio patched
  cPanel patched three critical vulnerabilities spanning privilege escalation, code execution and denial of service, with urgent upgrade guidance. Given cPanel's footprint on shared-hosting and SMB web infrastructure, patch coverage tends to lag — prioritize managed-hosting fleets.
  FreeBuf

Web Security Research

• ⚠️ Android adbd CVE-2026-0073 — zero-click nearby-attacker shell access
  Google disclosed a zero-click vulnerability in Android's adbd daemon allowing a nearby attacker to obtain a complete shell with no user interaction. High-value for the mobile attack-surface backlog alongside the recent Pixel 10 0-click chain; verify the June Android patch level rolled out across fleet devices.
  FreeBuf vuln portal

• ⚠️ ChatGPhish — ChatGPT web-summary renderer remains a phishing / indirect prompt-injection surface
  Permiso Security's disclosure keeps circulating: the chatgpt.com response renderer trusts Markdown links/images from third-party pages ChatGPT summarizes, auto-fetching attacker-hosted images and surfacing clickable links inside the trusted assistant UI — enabling indirect prompt injection, credential-phishing, and IP/User-Agent/Referer leakage. The summarizer is the attack surface.
  The Hacker News

AI Security

• ⚠️ Microsoft MDASH — 100+ AI agents surfaced 16 previously-unknown Windows vulnerabilities (4 critical)
  Microsoft's MDASH harness orchestrates 100+ specialized AI agents; in Windows testing it found 16 unknown vulnerabilities, four critical enough for remote control. Continuation of the AI-finds-bugs-at-scale arc alongside Anthropic's Mythos — the disclosure-to-exploit window keeps compressing, and defenders' patch cadence is the limiting factor.
  SecurityWeek

• 🛡 GitHub employee device compromised via malicious VS Code extension — internal repo access
  GitHub confirmed an employee device was compromised by a malicious Visual Studio Code extension, leading to unauthorized access to internal code repositories. Another data point in the IDE/extension-as-supply-chain pattern (cf. the QuickLens Chrome-extension compromise) — audit installed extensions and enforce allowlisting on developer endpoints.
  FreeBuf

Threat Intelligence

• Unit 42 — Iranian APT "Screening Serpens" expands espionage campaigns with AppDomainManager hijacking + new RAT
  Unit 42 details Screening Serpens targeting tech and defense sectors via AppDomainManager hijacking — a .NET technique that hijacks a legitimate app's execution flow by manipulating its config file, yielding arbitrary code execution before the host app even starts. The April RAT variant added an expanded dispatcher with 18 distinct opcodes, including file-chunking for stealthier exfiltration. Timeline spans late-2025 Western Europe targeting → Middle East, US, Israel, UAE samples through April 2026.
  Unit 42

• Carnival confirms ShinyHunters breach — 5,995,277 people affected, passport + driver's license data exposed
  Carnival Corporation confirmed a breach affecting nearly 6 million people (5,995,277 per the Maine filing). An attacker social-engineered an employee into granting access on April 14, then used a compromised account by April 22 to copy personal data — names, addresses, emails, phone numbers, DOB, driver's license and passport numbers — before being blocked. ShinyHunters claims 8.7M records and made data public in late April. Notifications began 5/27; two years of TransUnion credit monitoring offered to eligible US residents.
  Bleeping Computer · SecurityWeek · The Register · Help Net Security

Chinese-Language Community Picks

• FreeBuf / Yijing Lab and others: this week followed WP Maps Pro CVE-2026-8732 mass admin-account creation, Apache HTTP/2 CVE-2026-23918 double-free, Apache Flink CVE-2026-35194 SQLi, the cPanel CVE-2026-29201/2/3triple patch, and Android adbd CVE-2026-0073 0-click, among other English-sphere topics; the compromise of a GitHub employee's device via a malicious VS Code extension has heated up domestic CI/CD supply-chain discussion.
• The cPanel triple vulnerabilitydraws heavy attention from domestic site operators and hosting providers — cPanel holds a substantial share of shared hosting in China, so patch lag is a significant risk.
  FreeBuf· FreeBuf vulnerabilities channel

Ransomware Today

• ~30–40 new DLS posts in last 24hacross ~90 active leak sites. Most active: Qilin, Akira, DragonForce/Devman, The Gentlemen, Sinobi. Qilin sustains its multi-quarter lead with continued healthcare/legaltargeting; Akira active against manufacturing($244M cumulative proceeds, ~34% of IR engagements). Structural signals: DragonForce announced a "cartel" rebrandoffering affiliates 80% profit share + customizable encryptors + infrastructure (expect victim volume to rise), and the ransomware payment rate sits at an all-time-low 28%despite higher claimed volume. Watchlist hits today: Qilin × healthcare/legal, Akira × manufacturing.
• Full breakdown → intel/ransomware/daily/2026-06-02.html

Notable Breaches

• Carnival — 5,995,277 affected(ShinyHunters, see Threat Intel). A full-service collection agencydisclosed unauthorized access to customer data affecting 115,837 peoplein June. Multiple smaller breaches surfaced across sectors on 6/1.
  Bleeping Computer· BreachSense June 2026

---

AI Frontier

OpenAI

• Confidential IPO filing in preparation— reportedly working with Goldman Sachs and Morgan Stanley toward a confidential S-1 filing in the coming weeks, with a potential IPO as soon as September 2026. CNBC
• ChatGPhish— ChatGPT summary renderer abused for phishing + indirect prompt injection (see Web Security Research). The Hacker News

Anthropic

• Claude Opus 4.8 (5/28)— current flagship; stronger coding, agentic skills, and reasoning vs Opus 4.7, shipped in Claude Code with high-effort defaults. Releasebot
• Andrej Karpathy joins Anthropic— to work on AI-accelerated pre-training research, per recent reporting. LLM Stats updates
• Series H — $65B at $965B post-money (5/28)+ "Strengthening our model weight security"practices published. Anthropic· Model weight security
• Project Glasswing / Claude Mythos— 10,000+ high-severity flaws across widely-used software; first monthly update (5/22) remains the reference point. Code with Claude — Tokyo (Jun 5–6)upcoming. Anthropic

Google DeepMind / AI

• Gemini 3.5 Flash (cheaper enterprise tier)— Google pushing the lower-cost Flash model to win enterprise; Ultra cut to $200/mo, new $100/mo Developer tier. DeepMind licensed 20+ Contextual AI researchersin an ~$80–90M deal. HeyGoTrade recap
• Exposed client-side Google API keys can authenticate to Gemini— embedded keys (e.g., Maps) reusable to reach the Gemini assistant and access private data. Rotate and scope-restrict client-side keys. Bleeping Computer

🛡 = security-related

---

Failed Sources

• RansomLook API + RSS— blocked by sandbox egress allowlist (cowork-egress-blocked; host not on network allowlist). Ransomware section uses WebSearch aggregation of secondary trackers + quarterly reports; per-victim accuracy not guaranteed.
• Chinese sources (FreeBuf, 安全客, 先知社区)— direct feed fetch not available in current sandbox; coverage based on WebSearch summaries.

---

Sources used: see intel/sources.yaml