Rosetta Intel

Auto-generated · 29 sources configured (WebSearch mode) · 15 items selected
Window: past 24 hours (2026-05-21 → 2026-05-22)

In-the-Wild / New CISA KEV

🔴🔥⚠️ CISA adds 2 KEV entries (2026-05-21) — both AI-stack and EDR-stack pre-auth chains
- CVE-2025-34291— LangflowOrigin Validation Error (CVSS v4.0 9.4). Overly permissive CORS + SameSite=Nonerefresh-token cookie → cross-origin call to /refreshreturns valid tokens → full account takeover + RCEvia a single drive-by visit. Active exploitation observed since January 23, 2026, with attackers deploying the Flodric botnetthrough compromised Langflow instances. Federal due date 2026-06-11.
- CVE-2026-34926— Trend Micro Apex One on-premisedirectory traversal. Pre-auth-local attacker on the Apex One Server modifies the agent key table → arbitrary code pushed to all managed agents. Fix: ≥ 14.0.0.17079.
  CISA Alert· Windows Forum analysis· CrowdSec — Langflow exploitation· Obsidian Security — Langflow writeup
🔴⚠️ SonicWall Gen6 SSL-VPN — coordinated brute-force + MFA bypass campaign (5/20–5/21)
Threat actors exploiting CVE-2024-12802 (incomplete patch — MFA not enforced on UPN login format when SonicWall integrates with Active Directory). Full remediation requires six manual post-patch steps that are routinely missed, leaving "patched" appliances exposed. Attackers were in and out in 30–60 min: login → AD recon → credential reuse → logout, TTPs consistent with Akira ransomware operators. Gen6 reached end-of-life on April 16, 2026 — SonicWall no longer supports these. Replace, do not patch.
BleepingComputer · Cybersecurity Dive · SecurityAffairs · ReliaQuest threat spotlight

Critical Vulnerabilities

🔴⚠️ CVE-2026-42945 "NGINX Rift" — 18-year-old rewrite-module heap overflow (CVSS 9.2, public PoC + cross-request heap-shaping ASLR bypass)
Disclosed 5/13; weaponization continues. Inconsistent state in NGINX's two-pass script engine leaves the is_args flag set, mismatching buffer-size calc vs. copy-length. Unauthenticated heap-buffer overflow → worker crash (deterministic with ASLR on) → full unauth RCE when ASLR is off, ASLR-bypass chain when combined with a same-host LFI / arbitrary-file-read primitive. Affects every NGINX between 0.6.27 and 1.30.0, NGINX Plus R32–R36, and every downstream that ships ngx_http_script.c. Active exploitation observed in the wild this week.
The Hacker News — RCE chain · The Hacker News — in-the-wild exploitation · Akamai analysis · Picus Security writeup · HeroDevs — Ingress NGINX impact · BleepingComputer
⚠️ CVE-2026-46333 — Linux kernel (CVSS 5.5, disclosed 5/21). Capability misuse / local issue; lower-severity but continues the streak of kernel disclosures landing with PoC potential within the same day. Monitor distros for backports.
SecurityOnline coverage tracker · CVEFeed entry
⚠️ CVE-2026-5140 — Pardus Linux local privilege escalation chain
Critical LPE on a security-conscious distro; chain combines weak default permissions with a privileged helper. Flag because Pardus is used in Turkish government / education estates.

Vendor Advisories

Microsoft disrupts "Fox Tempest" Malware-Signing-as-a-Service — OpFauxSign (5/19, continued coverage 5/21)
Microsoft + FBI + Europol coordinated to seize signspace.cloud, take down hundreds of Azure tenants, and revoke 1,000+ code-signing certificates issued by Fox Tempest since May 2025. The MSaaS scheme let cybercriminals upload malware and receive certificates that signed it as Teams / AnyDesk / PuTTY / Webex. Tied to Rhysida ransomware (deployed by Vanilla Tempest), Oyster, Lumma Stealer, Vidar. The takedown is a meaningful blow to the supply-side of trusted-app impersonation campaigns.
Microsoft Security Blog · The Hacker News · SecurityWeek · The Record · Axios
Trend Micro Apex One on-premise — patch ≥ 14.0.0.17079. See KEV entry above; also patches CVE-2026-34926.
Trend Micro bulletin (Apex Central CV companion) · SecurityWeek

Web Security Research

"Worst-leak-I've-ever-seen" — CISA contractor exposes 844 MB of AWS GovCloud secrets on public GitHub (Krebs 5/19, continued fallout 5/21)
Contractor (Nightwing) maintained a "Private-CISA" repo that was actually public from Nov 13, 2025 to mid-May 2026 — 6+ months. GitGuardian's Guillaume Valadon flagged it on 5/15; CISA pulled it within 26 hours. Contents: plain-text passwords in a .CSV, AWS GovCloud tokens, SSH keys, Entra ID SAML certificates, internal CISA / DHS dev data. Commit log shows the contractor deliberately disabled GitHub secret scanning. Some exposed credentials remained active for 48 hours after takedown. Senator Hassan has formally pressed CISA for answers.
Krebs on Security · TechCrunch · The Register · Dark Reading · GitGuardian writeup · The Record — Senator Hassan letter
GitHub clarifies vector of the 3,800-repo internal breach (5/21): the trojanized Nx Console extension that compromised the GitHub employee was itself a downstream artifact of last week's TanStack npm supply-chain attack. The same TanStack token-rotation gap caused this week's separate Grafana data breach — a single un-rotated GitHub workflow token. The supply-chain "blast radius" of last week's TanStack incident is still widening.
BleepingComputer overview

AI Security

⚠️🛡 Langflow CVE-2025-34291 enters KEV (5/21) — first time CISA confirms in-the-wild exploitation of an AI-agent platform. The Flodric botnet is the active campaign. Anyone running self-hosted Langflow / similar AI workflow tools should treat them as internet-exposed identity surface. See KEV section for details.
GlobalSecurityMag — confirmed in-the-wild · Obsidian writeup
⚠️🛡 First confirmed AI-generated zero-day caught in the wild (GTIG 5/11, continued coverage 5/21)
Google Threat Intelligence Group published the first publicly attributed case of a criminal threat actor using an LLM to discover a previously unknown vulnerability — a 2FA bypass (hardcoded trust assumption, semantic logic flaw) in a popular open-source sysadmin tool. The exploit script bore strong LLM-generation signatures: hallucinated CVSS score, abundant educational docstrings, textbook Pythonic structure, ANSI color classes. Demonstrates the AI-as-weapon side of M-Trends 2026 — and Big Sleep / Mythos / Daybreak are the defensive counterweight.
GTIG / Google Cloud blog · TechRepublic analysis · Infosec.ge writeup
🛡 Anthropic [un]prompted 2026: Claude Mythos Preview autonomously generated 500+ validated zero-days
Confirmed during the conference talk and Project Glasswing landing — Mythos Preview demonstrated autonomous discover-and-exploit capability at meaningful scale: 500 zero-day exploits generated, validated, and ready-to-use, including a 17-year-old RCE in FreeBSD (CVE-2026-4747) giving unauthenticated root. Public access remains restricted; the 12-partner Project Glasswing consortium (AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMC, Linux Foundation, Microsoft, NVIDIA, Palo Alto, plus Anthropic) gets the find/fix workflow.
Anthropic Project Glasswing · red.anthropic.com — Mythos Preview · ·

Threat Intelligence

Unit 42 — CVE-2026-0300 PAN-OS Captive Portal pre-auth RCE under active state-sponsored exploitation
Disclosed 5/6 by Palo Alto Networks themselves; Unit 42 attributes ongoing exploitation to CL-STA-1132 (likely state-sponsored). Buffer overflow in User-ID Authentication Portal → unauthenticated root on PA-Series + VM-Series firewalls. Post-exploitation: AD enumeration, EarthWorm and ReverseSocks5 dropped. Patch immediately if you have an externally-reachable captive portal.
Unit 42 threat brief · The Register · The Hacker News
Unit 42 — "BlackFile" data-theft extortion (CL-CRI-1116) escalates to physical intimidation
Login via SSO with credentials phished via vishing → exfil customer records and internal comms → extortion without encryption (continues 2026 trend). New escalation: swatting of executives and employees as added pressure technique. Identity-as-initial-access continues to lead Unit 42's IR data — 65% of intrusions begin with identity-based techniques; data exfil now occurs in <1 hour in some cases (4× faster YoY).
Unit 42 May 2026 Threat Bulletin · Unit 42 Global IR Report 2026
Mandiant — "Another BRICKSTORM" stealthy backdoor expands into tech + legal sectors (continued)
China-nexus actor; modular .NET RAT family. Pair with Twill Typhoon FDMTP campaign (legitimate-binary DLL sideloading, CDN-impersonation C2: Yahoo / Apple lookalike domains) and GRIDTIDE (Google-Sheets-API-as-C2; 53 confirmed victims in 42 countries, 20+ suspected more). The 2026 China-nexus playbook is clearly "blend C2 into legitimate SaaS APIs."
Mandiant — Another BRICKSTORM · Disrupting GRIDTIDE

Chinese-Language Community Picks

Chinese-language sources today (FreeBuf / Xianzhi / Anquanke): JS rendering on Chinese sites is limited under WebSearch mode, so no specific new items for 5/21–5/22 were captured. Recommend checking the FreeBuf news feedand the Alibaba Cloud Vulnerability Databasedirectly. Likely follow-ups in the Chinese-language sphere today: NGINX Rift (CVE-2026-42945), the Langflow KEV entry, and the CISA leak incident.

Ransomware Today

Active leak-site posting continues at elevated tempo. New victims observed on 2026-05-21:

Qilin: Porter Yett (legal services), CJ Architects, CZ Collections
TheGentlemen: Grupo Pasquel
PEAR: ProFarm Group (ag-tech), Exchange Group, Fana Jewelry
LockBit (5.0 variant): Shottermill Junior School (k-12)

Watchlist hits: Qilin × 3, LockBit × 1 (education sector). The SonicWall Gen6 SSL-VPN brute-force campaign (above) is consistent with Akira's pre-positioning, expect Akira posts later in the week.

→ Full table: intel/ransomware/daily/2026-05-22.md

AI Frontier

OpenAI

OpenAI files confidential S-1 with SEC (5/22, reported)— targeting September 2026 listing at $852B–$1Tvaluation against $25B ARR. The IPO converts a private cap structure into public-market accountability for one of the largest pre-revenue-multiple AI bets in history.
OpenAI Deployment Company launched (continuing rollout)— separate entity to help enterprises build around OpenAI intelligence (system integration / vertical solutions).
OpenAI News· OpenAI Deployment Company

Anthropic

Claude Code for State and Local Governments — webinar 5/22
Demo-first session showing Claude Code mapping a realistic government codebase (COBOL eligibility module + .NET permitting), running test generation, and shipping a permit-status lookup service from scratch. FedRAMP-authorized deployment options discussed. Strategic move into a legacy-system modernization verticalwhere Anthropic is leading vs. OpenAI.
Anthropic webinar
Nearing first quarterly operating profit— Q2 2026 sales projected $10.9B; closes operating-loss gap dramatically faster than peers.
Claude Mythos Preview / Project Glasswing recap— see AI Security; 500+ validated zero-days demonstrate scale.
Anthropic News

Google DeepMind / AI

Gemini 3.5 Flash rolling out post-I/O 2026— frontier intelligence at lower latency; Ultra cut to $200/mo; Developer tier $100/mointroduced. Gemini MAU 900M(2× from May 2025).
DeepMind acqui-hires 20+ Contextual AI researchersin $80–90M licensing deal; CEO Douwe Kiela joining.
Gemini Spark— always-on consumer AI agent (general-purpose, background-running).
GTIG publishes first attributed AI-generated zero-day case— see AI Security.
Google blog — Gemini· Tom's Guide I/O 2026 recap

🛡 = security-relevant, archived under intel/ai-frontier/security/

Key Threads This Week

AI-agent platforms are now confirmed in-the-wild attack targets: Langflow CVE-2025-34291 entering CISA KEV (5/21) is the first time CISA has formally tagged an AI-workflow framework as actively exploited. Pair with Chroma ChromaToast(CVSS 10, last week). The AI infra surface is hot and attackers know it.
Old NGINX bugs > new NGINX bugs: an 18-year-old rewrite-module bug (CVE-2026-42945) shipping with public PoC + ASLR bypass + in-the-wild exploitation on the most-deployed web server on Earthis the highest-impact single bug this week. Patch + audit ngx_http_script.cdownstreams (every Ingress, every reverse proxy).
AI-as-weapon goes mainstream: GTIG's first publicly attributed AI-generated zero-day (Flodric / 2FA bypass) lands the same week Anthropic confirms Mythos Preview generated 500+ validated zero-days. Offense and defense are now both AI-powered, and only defense has consortium-scale capital behind it (Project Glasswing).
Last week's TanStack npm supply-chain attack is still expanding: GitHub-internal-breach AND Grafana breach both traced back to it this week. Single npm event → multiple infrastructure-tier compromises with multi-week blast radius. This is what 2026 supply-chain risk looks like.
CISA's own credential leak (Krebs)+ SonicWall Gen6 EOL exploitation = trust crisis at the edge of US federal cyber defense.

Failed Sources

This run used WebSearch mode (workspace web_fetch network allowlist blocks most security RSS hosts). Coverage is by topic, not raw feed scraping; some lower-priority feeds (Detectify Labs, Trail of Bits raw feed, 安全客, 先知社区) are not directly reflected. Items are cross-sourced from indexed news search.
RansomLook API not reachable from sandbox; ransomware section assembled from public leak-site mirrors and PurpleOps / ransomware.live aggregators.

Sources used: see intel/sources.yaml

Rosetta Daily · May 22, 2026