Rosetta Intel

Generated automatically · 31 sources configured (WebSearch aggregation + feeds) · 17 items selected
Window: past 24–48h (2026-06-12 → 2026-06-14). Sunday wrap-up. Today's throughline: the Oracle PeopleSoft zero-day (CVE-2026-35273, CVSS 9.8) has become the breach story of the month — ShinyHunters exploited it pre-patch to compromise ~300 instances across 100+ organizations, with 68% being universities (University of Nottingham alone: 454,600 student records). Still no full patch — mitigations only. A second weekend disclosure: "GreatXML", NightmareEclipse's new BitLocker bypass 0-day via Defender Offline Scan, dropped with full public details and no Microsoft fix. CrowdStrike's 2026 Global Threat Report lands the macro number that frames everything: 42% of vulnerabilities are now exploited before public disclosure, and eCrime breakout time has fallen to 29 minutes. AI security stays hot: fresh studies show GPT-5/Gemini agents fail direct prompt injection ~79% of the time.

In-the-Wild Exploitation (CISA KEV + active)

🔴🔥⚠️ Oracle PeopleSoft PeopleTools CVE-2026-35273 — CVSS 9.8 unauthenticated RCE, exploited as a 0-day by ShinyHunters; 100+ orgs hit, NO full patch
A remote, unauthenticated attacker with HTTP access can fully take over PeopleSoft Enterprise PeopleTools (8.61/8.62). Google/Mandiant observed exploitation May 27 → June 9 and notified 100+ organizations; ShinyHunters claims ~300 instances across 100+ orgs, 68% colleges/universities, exfiltrating student, financial-aid, immigration and health data. University of Nottingham alone: 454,600 records published. Oracle has shipped emergency mitigations only — patch still pending. Action: apply Oracle's mitigations now; hunt PeopleSoft web tiers for webshells/anomalous auth back to 5/27; assume data theft if exposed.
The Hacker News · BleepingComputer · SecurityWeek
⚠️ "GreatXML" BitLocker bypass 0-day — full public disclosure, no Microsoft patch
Researcher NightmareEclipse published a working PoC (built in ~4 hours) that bypasses BitLocker by abusing the Windows Recovery Environment + the unattend.xml config used by Defender Offline Scan. Any machine that has ever run a Defender Offline Scan becomes permanently susceptible — no login required. It's a post-compromise technique (needs admin to write the recovery partition), but it means full disk decryption after an attacker lands admin. This is the 8th tool from the Nightmare-Eclipse cluster in ~10 weeks and the 2nd BitLocker bypass after "YellowKey". Action: enforce pre-boot auth (TPM+PIN); restrict admin/recovery-partition write; monitor WinRE/unattend.xml tampering.
SecurityWeek · Cybersecurity News · TechRepublic
🔥⚠️ Carryover KEV trio — June 23 federal deadline still live
The Chrome V8 (CVE-2026-11645), Cisco SD-WAN Manager (CVE-2026-20245, no patch) and Arista EOS (CVE-2026-7473, won't-fix) KEV entries from 6/9 remain due June 23. Action: confirm Chrome auto-update landed fleet-wide; lock down SD-WAN management plane; apply upstream ACLs for the Arista decap flaw.
CISA KEV

Critical Vulnerabilities

⚠️ Nightmare-Eclipse cluster keeps shipping post-patch Windows exploits
GreatXML follows last week's RoguePlanet (Defender LPE → SYSTEM) and the BitUnlocker downgrade attack — a sustained stream of public Windows offensive tooling outpacing Microsoft's fixes, much of it released amid an open disclosure dispute with MSRC. Action: treat fully-patched Win11/10 as still exposed to local privesc + BitLocker bypass; prioritize physical-access and admin-containment controls.
SecurityWeek
🔴 Fortinet June batch — FortiOS / FortiSandbox / FortiProxy fixes (incl. critical FortiSandbox flaw)
Fortinet shipped updates across FortiOS, FortiSandbox and FortiProxy alongside the June Patch-Tuesday wave, including a critical FortiSandbox vulnerability. Edge/security appliances remain the #1 ransomware initial-access surface. Action: patch internet-facing Forti appliances on priority; review admin exposure.
Security Affairs · SANS NewsBites

Vendor Advisories

Oracle— out-of-cycle CVE-2026-35273(PeopleTools, CVSS 9.8) emergency mitigations; full patch pending. SecurityWeek
Microsoft— no patchfor GreatXML BitLocker bypass; researcher disclosure dispute ongoing. TechRepublic
Cisco— 10 advisories / 15 issues this cycle incl. 4 critical and a Unified CM PoC, on top of the SD-WAN KEV 0-day. SANS NewsBites
Fortinet— FortiOS / FortiSandbox / FortiProxy fixes incl. critical FortiSandbox. Security Affairs

Web Security Research

🛡 "When prompts become shells" — RCE in AI agent frameworks (Microsoft Security)
Microsoft details how prompt injection crosses from a content problem into a code-execution primitive once an LLM is wired to tools: in agentic frameworks and MCP servers, injected instructions can reach shell/tool execution. Action: sandbox agent tool execution; treat MCP tool descriptions and retrieved content as untrusted; enforce least-privilege on connectors.
Microsoft Security Blog
PortSwigger Research — Burp SSRF testing workflow refreshed (6/8) with updated guidance on private-IP enumeration and blind/out-of-band SSRF via Collaborator. Action: fold the updated playbook into AppSec regression tests for cloud-metadata and internal-service SSRF.
PortSwigger

AI Security

🔴⚠️🛡 Prompt injection remains unsolved — GPT-5/Gemini agents fail direct injection ~79% of the time
New research (covered at Infosecurity Europe and by OWASP) shows leading agents still can't reliably resist prompt injection: direct attacks succeed >79% of the time, and hidden/web-embedded (indirect) injection routinely hijacks behavior. Root cause is architectural — models process system prompt, user input and retrieved content in one token stream with no privilege boundary. Action: assume any tool-wired agent is injectable; gate tool/connector scopes; log and review model I/O.
Decrypt · Help Net Security · Infosecurity
🛡 CrowdStrike: GenAI tools abused at 90+ organizations
The 2026 Global Threat Report documents adversaries injecting malicious prompts into legitimate GenAI tools at 90+ orgs to generate credential- and crypto-stealing commands — prompt injection operationalized at scale. Action: monitor enterprise GenAI usage; restrict agent actions; alert on AI-generated offensive command patterns.
CrowdStrike

Threat Intelligence

🛡 CrowdStrike 2026 Global Threat Report — 42% of CVEs exploited before disclosure; breakout time 29 minutes (fastest ever: 27 seconds)
Adversaries weaponized zero-days for initial access, RCE and privesc, with 42% of vulnerabilities exploited pre-disclosure. Average eCrime breakout time fell to 29 minutes (65% faster than 2024), cloud-conscious intrusions rose 37%, and state-nexus cloud targeting jumped 266%. Action: assume minutes-not-days; prioritize identity + cloud detection; patch on threat-intel signals, not just CVE publication.
CrowdStrike
🛡 Law enforcement disrupts "AudiA6" crypto-laundering pipeline used by ransomware gangs
European authorities took down AudiA6, a cryptocurrency-laundering service that washed "hundreds of millions" in illicit profits for ransomware operators and cybercrime networks. Action: track downstream effects on affiliate payout flows; expect short-term operational friction for tied groups.
PrivacyGuides — breach roundup

AI Frontier

OpenAI

Quiet weekend on launches; GPT-5.5 Instantremains the default ChatGPT model. Post-confidential-IPO-filing positioning continues against Anthropic's earlier filing.

Anthropic

Project Glasswing expands to ~150 organizationswith access to the Claude Mythoscyber preview; Anthropic says early partners have surfaced 10,000+ high-risk vulnerabilitiesvia the program — a defensive counterweight to AI-accelerated offense.

Google DeepMind / AI

Gemini Spark— a general-purpose AI agent that reasons across connected apps — is rolling out in beta to trusted testers and AI Ultra subscribers; the "Omni" world modelwas also unveiled. Two Gemini preview models are deprecated, closing June 25. Gemini 3.5 Flashremains the default in the Gemini app and Search AI Mode.

🛡 = security-relevant

Chinese-Language Community Picks

FreeBuf — Pwn2Own Berlin 2026 saw live demonstrations of multiple 0-days against NVIDIA, Docker, Linux KVM and PyTorch; separately, a 0-day RCE targeting Microsoft Office is being offered for sale on hacker forums. Action: track the patch cadence for the container / virtualisation / AI framework stack; stay alert to Office document phishing.
FreeBuf/CSDN
Anthropic's Project Glasswing expands to roughly 150 organisations (per FreeBuf), opening the Claude Mythos cybersecurity preview; the company says early partners have used it to find 10,000+ high-severity vulnerabilities — a sample of AI applied at scale on the defensive side.
Tencent Cloud Developer Community

Ransomware Today

Leak-site activity stayed heavy into the weekend. On June 11, trackers logged 39 new leaks in 24h, led by Qilin (18 victims, concentrated in manufacturing and energy), followed by DragonForce (7), M3RXDLS (6) and DireWolf (4); top sectors were Real Estate, Manufacturing and Construction/Engineering, hitting the US and UAE. Qilin remains the most active operation of 2026 (~500 victims YTD, ~1,500 lifetime). Separately, the ShinyHunters PeopleSoft campaign (see KEV section) is the dominant data-theft story — extortion of universities is expected to follow. Watchlist hits: groups qilin, dragonforce; sectors manufacturing, energy. → full victim table

Other Notable Incidents

SoFi Hong Kong— customer data exposed via a breached third-party vendordatabase. PrivacyGuides
Tchap (French civil-service messenger)— social-engineered account let an attacker scrape 13.5GB(560,000 messages, 73,000+ accounts) from the French tax authority and other agencies. PrivacyGuides
Pwn2Own Berlin 2026 (via FreeBuf)— fresh 0-days demonstrated against NVIDIA, Docker, Linux KVM and PyTorch; an Office 0-day RCE is reportedly being sold on hacker forums. FreeBuf/CSDN

Failed / Degraded Sources

RansomLook API(/api/posts?days=1) and RSS, plus ransomware.live API— empty/blocked from the sandbox (egress/provenance restrictions). Ransomware section reconstructed via WebSearch (Ransom-DB / Bitdefender / MOXFIVE trackers); counts are a lower bound, not a full leak-site snapshot.
Direct feed fetch (WebFetch) not used this run — all categories sourced via WebSearch aggregation. Per-feed entry timestamps approximate; several items span the 6/10–6/13 window still in scope on a weekend run.

Sources used: see intel/sources.yaml

Rosetta Daily · Jun 14, 2026