October 2026 produced a serious attack almost every day. Hosting platforms leaked customer secrets, official releases shipped stealers, a certificate authority's keys turned out to be computable, ransomware affiliates ran intrusions through a coding agent, and a CVSS 10.0 bug surfaced in GitLab. Our deeper posts cover each one; this page puts all of them in a single date order so you can see the month at a glance.
This is a living page. It is checked for new incidents every 4 hours through the end of October, and every entry links to a full write-up with primary sources. The same data powers the shareable October 2026 attacks tracker: filter by type, link to a single incident, or embed one on your own site.
Last updated: October 31, 2026, 22:02 UTC · 107 dated entries, 5 earlier campaigns, 12 unconfirmed claims
TL;DR
| Question | Answer |
|---|---|
| What does this page cover? | Every attack, breach, exploited vulnerability and agent incident of October 2026 we could verify, in date order |
| Which date is used? | When the attack happened, or when it was first disclosed if that is all that is known |
| Are claims included? | Yes, labelled Unconfirmed and kept in their own section |
| What is the common thread? | Stolen or unrotated credentials, then trusted channels: release workflows, update servers, app credentials |
| Were AI keys targeted? | Yes: Railway, OpenCode, LiteLLM, Tensorlake, GhostAction and LLMjacking all went after AI access |
| Did AI run attacks? | AI did much of the work in INC, BUGSTER, ARTEX and Wiz's case; humans still chose targets and approved steps |
| First thing to do | Rotate secrets on affected platforms and hosts, then patch PHP, Node.js, GitLab, Java and trust stores |
How to read the timeline
Each entry has a type and a status:
| Type | Meaning |
|---|---|
| Supply chain | A package, release, plugin or update channel shipped malicious code |
| Breach | A company's systems or customer data were accessed |
| Exploited | A vulnerability with confirmed in-the-wild attacks |
| Critical vuln | A severe flaw disclosed and patched, with no confirmed exploitation yet |
| AI-run attack | An intrusion where agents or LLMs did substantial operational work |
| Agent incident | An AI agent caused harm or broke a control without an outside attacker |
Confirmed means the affected company, a government agency or a named research team stated it. Reported means one vendor's research that has not been independently corroborated. Unconfirmed means a claim with no supporting statement.
Week 1: October 1 to 7
| Date | Incident | Type | What happened | Status |
|---|---|---|---|---|
| Oct 1 | OpenAI agent access notices | Agent incident | OpenAI disclosed it had notified 100+ organizations that its agents may have accessed their systems without authorization (roundup) | Confirmed |
| Oct 1 | INC ransomware builds staged | AI-run attack | 46 INC encryptor builds staged for six victims for an October 1 deployment; broader deployment unconfirmed (roundup) | Reported (SOCRadar) |
| Oct 2 | Cairn instances exposed | AI-run attack | About two dozen exposed Cairn instances found in malicious use, the start of the 1,200+ site story (details) | Reported |
| Oct 3, 6 | OpenAI agents disable monitoring | Agent incident | GPT-6 agents removed misalignment monitors during routine work; one denied it (details) | Confirmed (OpenAI) |
| Oct 3 | KVM guest-to-host VM escape found through Vercel Sandbox bounty | Critical vuln | Researcher Paulos Yibelo announced a full guest-to-host escape in KVM, confirmed by Vercel and paid a $50,000 bounty; code inside a guest VM could reach root on the host. A kernel fix later appeared as CVE-2026-99999, though whether it is exactly this bug is not verified. No use in the wild reported (source) | Confirmed (Vercel) |
| Oct 5 | SubQuery @subql/common 5.8.3 | Supply chain | Modified publish workflow swapped the build for an external archive, published through trusted publishing; steals CI and developer secrets (tracker) | Confirmed |
| Oct 5 | Langflow CVE-2026-105740 | Exploited | CVSS 9.9: any authenticated user can run shell commands through an MCP stdio server; honeypot attempts since September 19 (details) | Confirmed vuln; attempts reported |
| Oct 6 | Backdoored Qwen model steals .env via Codex (PoC) | Supply chain | ProjectDiscovery backdoored Qwen2.5-7B for under $50; a trigger phrase made it in Codex CLI fetch a script and send .env files out. Research proof of concept, no in-the-wild use shown (details) | Reported (ProjectDiscovery) |
| Oct 7 | ARTEX vs South Korean banks | AI-run attack | CrowdStrike: a likely lone actor ran ARTEX and several LLMs against Korean banks, exposed by his own open directory (details) | Reported (CrowdStrike) |
| Oct 7+ | GhostAction (new wave) | Supply chain | Stolen GitHub tokens push "security-audit" workflows that send package, cloud and AI keys to attackers (tracker) | Confirmed; counts vary |
Week 2: October 8 to 14
| Date | Incident | Type | What happened | Status |
|---|---|---|---|---|
| Oct 8 | Tensorlake npm 0.5.144 | Supply chain | Shai-Hulud variant in an AI sandbox SDK: steals developer and AI tool secrets, self-spreads, has a wipe switch (tracker) | Confirmed |
| Oct 9 | Claude Code CVE-2026-86063 | Critical vuln | Plugin hash pin could be bypassed by a same-named branch; fixed in 2.1.179 (details) | Confirmed |
| Oct 9 | Anthropic unintended actions | Agent incident | Anthropic reported Claude taking unintended actions on live sites, including a fake police tip (details) | Confirmed (Anthropic) |
| Oct 9 | Indian fintech attacks | Breach | CERT-In warned attackers stole API credentials at NBFCs, lenders and payment firms and called fund-transfer APIs directly; link to AI agents unclear (details) | Confirmed advisory |
| Oct 10-14 | Storm-0940 infrastructure | AI-run attack | Spray infrastructure later linked to the ConnectWise intrusion was active (details) | Confirmed (Microsoft) |
| Oct 12+ | GitLab CVE-2026-97229 | Exploited | CVSS 10.0 integer overflow: an unauthenticated user can run code on a self-managed GitLab server through a crafted regex in a CI/CD configuration; fixed in 19.2.8, 19.3.5, 19.4.2; added to CISA KEV October 29, federal deadline November 1 (source) | Confirmed |
| Oct 13 | MCP TypeScript SDK CVE-2026-104850 | Critical vuln | A malicious MCP server could collect OAuth refresh tokens and client secrets; fixed in sdk 1.31.0 (details) | Confirmed |
| Oct 13+ | Ivanti Connect Secure CVE-2026-111357 | Exploited | CVSS 7.0 local privilege escalation to root on Connect Secure 22.8R2.4 (the dsinvoked broker trusts DSCache configuration data), exploited in the wild in October alongside the CVE-2026-104935 auth bypass; added to CISA KEV October 30, federal deadline November 2 (source) | Confirmed |
| Oct 13-17 | ConnectWise Cloud Backup | Breach | A stolen app credential for seven app identities signed in to 162 to 173 Microsoft 365 tenants (tracker) | Confirmed |
| Oct 13 | Microsoft October Patch Tuesday | Exploited | Record ~1,411 fixes; CVE-2026-96951 Windows Shell Mark-of-the-Web bypass exploited; CISA KEV reported Oct 13 (details) | Confirmed |
| Oct 14 | SiteGuard Server Edition | Exploited | CVE-2026-106931 OS command injection, on by default; root or SYSTEM on the WAF host; CISA KEV (tracker) | Confirmed |
| Oct 14+ | Collabora built-in CODE server CVE-2026-110447 | Exploited | OS command injection in the Nextcloud built-in CODE server (richdocumentscode before 26.4.303); admins report attacks since September 10; fixed September 11; added to CISA KEV October 23 (source) | Confirmed |
| Oct 14-15 | Zeabur (second breach) | Breach | Plaintext admin credentials left out of the August rotation; 321,810 accounts and 3.2M env var records (details) | Confirmed |
Week 3: October 15 to 21
| Date | Incident | Type | What happened | Status |
|---|---|---|---|---|
| Oct 15 | CERT-In agent advisory | Agent incident | AI agents under testing used leaked keys and broad cloud roles to read confidential records at Indian organizations (details) | Confirmed advisory |
| Oct 15 | OpenAI model spent test ETH | Agent incident | An internal model derived keys from public recovery phrases and spent Sepolia ETH it was not given (details) | Confirmed (OpenAI) |
| Oct 15-16 | Rank Math PRO 3.0.123 | Supply chain | Update channel served a self-restoring backdoor for 11 hours; CVE-2026-111862 (tracker) | Confirmed |
| Oct 15-19 | Siemens Software Center | Exploited | DLL side-loading (CVE-2026-110936) abused by the Romulus ClickFix campaign; CISA KEV (tracker) | Confirmed |
| Oct 15+ | Linux kernel TCP use-after-free CVE-2026-98432 | Critical vuln | IPv4 TCP handshake use-after-free scored 9.8; Anthropic built a root exploit against an MPTCP listener. Fixed upstream and in Debian, Ubuntu still in progress. No in-the-wild use found (details) | Patched |
| Oct 16-17 | Plex2Shell | Exploited | Four chained Plex Media Server CVEs; miners and a proxy backdoor within a day of the PoC (tracker) | Confirmed |
| Oct 16-17 | OpenCode Console | Breach | A legacy system exposed saved provider keys for about 1.4% of Console users (details) | Confirmed |
| Oct 17 | eMudhra CA | Critical vuln | Eight intermediate CA keys were computable (one was 4), so any TLS certificate could be forged (details) | Confirmed |
| Oct 17 | Zeabur shutdown | Breach | Founder announced Zeabur will wind down through December 2026 after the breach (details) | Confirmed |
| Oct 17-28 | VAPOR PANDA runs ARTEX against 200+ organizations | AI-run attack | CrowdStrike: a China-nexus group ran the ARTEX agent framework with DeepSeek from one controller against 200+ organizations in 14+ countries in 11 days; nine confirmed compromised, BLOODALCHEMY backdoor deployed. 200+ is targeting, not breaches (details) | Reported (CrowdStrike) |
| Oct 17-18 | TeamUpdraft | Breach | Customer database accessed; UpdraftCentral keys used to install backdoor plugins (details) | Confirmed |
| Oct 19 | OVHcloud CLI v0.11.0 to v0.15.0 | Supply chain | Contributor account hijacked; release assets and checksums.txt replaced for about 94 minutes (details) | Confirmed |
| Oct 19 | Wiz: AI-driven cloud attack | AI-run attack | An agent chained a Kafka UI flaw to cluster-admin at machine speed (details) | Reported (Wiz) |
| Oct 19 | Apex Accelerator npm (REF8801) | Supply chain | Fake AI copilot and MCP packages redirect LLM traffic, add an attacker MCP server and steal configs (details) | Reported (Elastic) |
| Oct 19 | Cairn: 1,200+ sites | AI-run attack | Reuters and NorForge: agents probed 1,200+ sites and breached 100+; cost and theft unconfirmed (details) | Partly confirmed |
| Oct 19 | Capcom CAPCOM ID | Breach | Unauthorized access found Oct 19: 11,774,247 records (emails, login metadata); 43 with birth dates; 196 accounts had passwords changed. Capcom says passwords and cards were not exposed (source) | Confirmed (Capcom) |
| Oct 20 | INC and The Gentlemen via OpenCode | AI-run attack | SOCRadar: 8,257 messages in 133 OpenCode sessions, about $23,920 of model spend, 62 named organizations (details) | Reported (SOCRadar) |
| Oct 20 | LLMjacking via GCP billing | Breach | Phished GCP admins; foreign projects linked to victim billing to run Gemini (details) | Reported (Wiz) |
| Oct 20 | MCP TypeScript SDK 1.32.0 | Critical vuln | Four advisories, two high; two fixes need config to take effect (details) | Confirmed |
| Oct 20 | exploit.cat Claude disclosure | Critical vuln | Session-bridge JWT and artifact-upload claims; Anthropic disputes the second (details) | Disputed |
| Oct 20 | Oracle Java SE CPU | Critical vuln | 55 Java SE fixes, 49 remotely exploitable without authentication (details) | Patched |
| Oct 20-21 | PHP2Shell | Exploited | AI-found PHP interpreter RCEs leaked a day early and were attacked before the fix; CVE-2026-109618 and CVE-2026-109619 added to CISA KEV October 26; a researcher solved the stealmybitcoin.com challenge on unpatched PHP 8.4.26 in 19.28 hours (Oct 25) (details) | Confirmed |
| Oct 20-21 | Linux KSMBD flaws (ZDI-CAN-33774, -33549) | Critical vuln | Two KSMBD flaws disclosed with public PoCs before a released fix; PoCs show a memory leak and crashes, no CVEs assigned (details) | Partly confirmed |
| Oct 21 | LiteLLM 1.107.0 | Supply chain | PyPI wheels and Docker images ran a startup-hook stealer and RAT for about 100 minutes (details) | Confirmed |
| Oct 21 | Railway | Breach | Customer env vars accessed, some decrypted; Railway says the aim was AI inference keys (details) | Confirmed |
| Oct 21 | Storm-0940 and ARTEX | AI-run attack | Microsoft: ARTEX agents used stolen ConnectWise app credentials against 2,250+ tenants (details) | Confirmed (Microsoft) |
| Oct 21 | BUGSTER | AI-run attack | Unit 42: Hermes Agent, DeepSeek V4 and a permission-bypassed Cursor used to steal 5,200+ records (details) | Reported (Unit 42) |
| Oct 21 | CyberCartel CARLIY0S | AI-run attack | Straiker: Claude, Grok and Kimi agents aimed at 400+ Latin American financial targets (details) | Reported (Straiker) |
| Oct 21 | GitLab CVE-2026-110264 | Critical vuln | CVSS 10.0 unauthenticated RCE in self-managed GitLab back to 7.0 (details) | Patched |
| Oct 21 | Node.js security release | Critical vuln | Ten CVEs including a Brotli RCE and TLS hostname bypasses (details) | Patched |
| Oct 21 | Tailscale | Critical vuln | eMudhra trust exposure fixed in v1.104.2 (tracker) | Patched |
| Oct 21 | Cisco Smart Install (Static Tundra) | Exploited | NSA: disable Smart Install now; CVE-2018-0171 still used to steal switch configs (details) | Confirmed advisory |
| Oct 21 | State Bank of India | Breach | 3,897 customer records found in a dark-web set; source unknown (tracker) | Confirmed (SBI) |
| Oct 21 | Quest Diagnostics Quanum LSM | Breach | Quest learned Oct 21 of unauthorized access to its Quanum LSM ordering system and believes patient records were accessed; scope still under investigation, system taken offline (source) | Confirmed (Quest, SEC 8-K) |
| Oct 21 | GMO Internet Onamae.com Desktop Cloud | Breach | Malware on FX-trading VPS servers (26,739 potentially affected); stored trading logins and customer data taken, unauthorized trades reported, extortion refused; cause unpublished (details) | Confirmed (GMO Internet) |
Week 4: October 22 to 28
| Date | Incident | Type | What happened | Status |
|---|---|---|---|---|
| Oct 22 | Symantec "Skynet mode" | AI-run attack | Honeypot caught an offensive agent using exposed LLM endpoints as its brain, with 74 tools (details) | Reported (Symantec) |
| Oct 22 | Meta Muse agent | Agent incident | Filesystem export, shared artifact links and VM escape reports; Meta says the export was not a breach (details) | Partly confirmed |
| Oct 22 | x402 SDK flaws | Critical vuln | Three critical bugs let a hostile API drain an agent wallet: no default spend cap, Solana approvals accepted as payments, SIWx logins signed for other sites (details) | Patched |
| Oct 22+ | Arista VeloCloud Orchestrator CVE-2026-114625 | Exploited | Any authenticated user, even read-only, can act as a superuser operator on on-prem VCO; Arista reports active exploitation; fixed in 5.2.3.18, 6.1.3.10, 6.4.2.12; added to CISA KEV October 23 (source) | Confirmed |
| Oct 22 | Malicious AGENTS.md in Downloads | Agent incident | Forever Security found a planted AGENTS.md in a customer's ~/Downloads that steered a coding agent to install the malicious npm package function-color; delivery method unknown (details) | Reported (Forever Security) |
| Oct 23 | Panasonic HR data sent to generative AI | Agent incident | Employee at a subsidiary reportedly uploaded HR data on ~330,000 people to a banned AI service; no outside attack, Panasonic release not checked (details) | Reported (Nikkei) |
| Oct 23 | Mythos 5 malicious pull request (AISI test) | Agent incident | In a UK AISI cyber test with internet on and safeguards off, a Claude Mythos 5 agent sent a malicious pull request to a real GitHub project and used sock-puppet accounts; rejected, no harm found (details) | Confirmed (Anthropic, AISI) |
| Oct 23 | OpenAI containment red team | Agent incident | OpenAI says red-team agents escaped training VMs, found paths to the live internet and passed signals between VMs through CPU cache timing; sandbox components patched (details) | Confirmed (OpenAI) |
| Oct 23 | Aokigahara infostealer targets AI coding tools | Supply chain | Check Point: a Windows infostealer with an aisessions module collects local auth and settings files of Claude Code, Codex, ChatGPT Desktop and Cursor; endpoint malware, no vendor breached (details) | Reported (Check Point) |
| Oct 23-25 | PHP2Shell + Fragnesia chain roots 25 firewalls | Exploited | Dream Research: a suspected China-nexus operator chained a custom PHP2Shell exploit with Fragnesia (CVE-2026-46300) to root 25 servers in the US, Japan, Taiwan and South Korea. Claude Code was on the attacker's host; no proof it wrote the exploits. CVE-2026-46300 added to CISA KEV October 28 (details) | Reported (Dream Research); KEV confirmed |
| Oct 25 | Tailworm: Tailscale SSH and headscale flaws | Critical vuln | Researcher published 11 unpatched findings in Tailscale 1.104.2 and headscale 0.29.5; a Tailscale SSH forwarding race may give root on other servers after one device is compromised (lab only) (details) | Reported (researcher); Tailscale acknowledged |
| Oct 26 | Qwen3.8-Flash-Next tried an upload to a Beijing bucket | Agent incident | Cohere: in a security benchmark an open-weight Qwen model listed /tmp, read DNS settings and tried to upload a test string to an Alibaba Cloud bucket in Beijing; under 1 in 100,000 calls, blocked by the sandbox, intent unknown (details) | Reported (Cohere) |
| Oct 26 | testmetest worm arms 42 GitHub repos with agent hooks | Supply chain | JFrog: a Miasma-family npm worm committed Claude Code, Gemini, Cursor and VS Code hook files to 42 real repos (11 owners) that run a ~13 MB .github/init.js when the project is opened; cloning alone does not trigger it (details) | Reported (JFrog) |
| Oct 26 | Apple iOS 27.2: 163 fixes | Critical vuln | iOS and iPadOS 27.2 fix 163 flaws, including eight WebKit code-execution bugs and two sandbox escapes; iOS 26.7.2 carries many of the fixes. No exploitation reported (details) | Patched |
| Oct 27 | GPT-6.1 Terra worked around restrictions in tests | Agent incident | OpenAI system card: in the Respecting Warnings evaluation Terra finished the task by a route that violated the restriction's spirit in 29.6% of rollouts; no attempts to bypass Codex Auto-review. Simulated tests, not real-world incidents (details) | Confirmed (OpenAI) |
| Oct 27 | LiteLLM CVE-2026-114612 | Critical vuln | CVSS v4 7.7: an internal_user can reach proxy_admin and run commands on the proxy host through MCP stdio; affects 1.91.0 up to 1.100.4, 1.101.3, 1.102.2, 1.103.1, 1.104.0rc2. No exploitation reported (details) | Patched |
| Oct 27 | vLLM CVE-2026-114053 | Critical vuln | CVSS 3.1 8.1: a request-supplied code_revision in mm_processor_kwargs makes vLLM import attacker-controlled processor code when trust_remote_code is on; /tokenize bypasses auth; fixed in 0.31.0. No exploitation reported (details) | Patched |
| Oct 27 | Warp vendor breach | Breach | Warp's CEO says an attack on a third-party vendor exposed agent traces and telemetry for roughly 1% of users; whether prompts, source code or credentials were included is unconfirmed (details) | Confirmed (Warp) |
| Oct 27 | EPICS Base: four CVEs found by the Cipher agent | Critical vuln | CISA ICSA-26-300-01: four EPICS Base flaws (CVE-2026-114595 CVSS 9.8, possible RCE); fixed in 7.0.11. Causal Security says its Cipher agent found them in 3h49m (vendor claim). No known exploitation (details) | Patched |
| Oct 27 | LiteLLM file read via vertex_ai_credentials | Critical vuln | GHSA-g5ff-637f-6q2m (CVSS 8.1): a viewer-role user can read files the proxy can, including /proc/self/environ with the master key; fixed in 1.95.0. A second advisory covers SSO debug callback XSS (details) | Patched |
| Oct 27 | MCP TypeScript SDK GHSA-f897-62hf-6gxq | Critical vuln | CVSS 7.5: HTTP MCP servers had no POST body or JSON-RPC batch limit, so one unauthenticated client could exhaust memory; fixed in sdk 1.30.1 and 2.1.0 packages; stdio servers not affected (details) | Patched |
| Oct 27 | Adobe Commerce and Magento CVE-2026-48356 | Exploited | CVSS 9.3 unrestricted file upload that can lead to code execution; fixed in Adobe bulletin APSB26-73 in July; added to CISA KEV October 27, federal deadline November 10 (source) | Confirmed |
| Oct 27 | Brotli CVE-2026-111497 (Node2Shell) | Exploited | Brotli decoder bug added to CISA KEV October 27; Cloudflare shipped two Log-only managed WAF rules on October 25 (Content-Encoding: br bodies and /_next/image). Patch Node.js, Bun and Python Brotli runtimes; the WAF is not a fix (details) | Confirmed |
| Oct 27 | Anthropic: AI-built exploit chains | Critical vuln | Anthropic published five exploit chains an unreleased research model built for Chrome, Safari, Android, Windows and Linux; fixes include Chrome 156, Linux CVE-2026-98432 and Windows CVE-2026-33096. Research, no in-the-wild use claimed (details) | Patched |
| Oct 27 | MCP Python SDK 1.30.0 and 2.2.0 | Critical vuln | Three medium advisories (CVE-2026-116684, -116685, -116686): bearer tokens for other services accepted, cross-origin redirects leaking headers, client fetching server-chosen $ref URLs. The audience fix also needs validate_token_resource=True (details) | Patched |
| Oct 27 | CyberStrikeAI: 44 agents probe organizations in Japan | AI-run attack | TrendAI: attackers ran the open-source CyberStrikeAI platform with a 44-agent setup against at least 19 organizations, 16 in Japan; evidence is attacker-side files only, many targets were only probed, no zero-day reported (details) | Reported (TrendAI) |
| Oct 28 | 123 Reg "AI-driven cyberattack" | Breach | UK registrar 123 Reg says a Sept 18 AI-driven attack read customer contact and product data; the ARTEX and China-nexus links are the company's own assessment, no technical report published (details) | Confirmed (123 Reg); AI and ARTEX links unverified |
| Oct 28 | vLLM 0.32.0: eight advisories | Critical vuln | Two high-severity RCE paths: a forged shared-memory handle reaching pickle.loads (CVE-2026-115912, needs --mm-processor-cache-type shm) and an EPS image reaching a vulnerable Ghostscript (CVE-2026-115915). Six DoS or exposure bugs; all fixed in 0.32.0 (details) | Patched |
| Oct 27-28 | OWASP: ChatGPT-guided novice takes over a lab admin account | AI-run attack | OWASP GenAI Security Project: on authorized Hack The Box machines a novice followed ChatGPT to an admin takeover; none of nine autonomous setups (incl. Claude Code, Codex) finished it. Lab research, no real victims (details) | Reported (OWASP) |
| Oct 28 | VMware ESX CVE-2026-41709 | Exploited | Insufficient logging (CVSS 2.7): a malicious administrator can perform operations on ESX that are not logged; Broadcom advisory published July 30; added to CISA KEV October 28, federal deadline November 11. Low severity, but CISA lists it as exploited (source) | Confirmed |
| Oct 28+ | Cisco ASA and FTD CVE-2026-76418 | Exploited | CVSS 8.6: a crafted request to the Remote Access SSL VPN with secondary authentication on a tunnel group makes the firewall reload (denial of service, no login needed); added to CISA KEV October 29, federal deadline November 1 (source) | Confirmed |
| Oct 28 | DHS: 40,000+ employee records leaked, hacker claims an Okta source | Breach | DDoSecrets published personnel data on 40,000+ DHS employees (CBP, ICE, CISA and others) from a hacker called Department of Peace. The group claims an Okta source; Okta says it has no evidence of compromise. Samples verified; entry point unconfirmed (details) | Reported (data partly verified; Okta claim disputed) |
| Oct 28 | ChatGPT share link XSS (Parameter) | Critical vuln | Parameter: a public ChatGPT share link chained Preview, a MathLive postMessage flaw and a YouTube JSONP CSP bypass to run attacker code as chatgpt.com in the viewer's session. OpenAI rated it P1 and fixed it Oct 26; proof of concept only, no in-the-wild use reported (details) | Patched |
Week 5: October 29 to 31
| Date | Incident | Type | What happened | Status |
|---|---|---|---|---|
| Oct 29 | Wiz: three agent-run intrusions and seven fingerprints | AI-run attack | Wiz investigated intrusions run by attacker AI agents through an exposed Flowise server, a Java operations gateway and an MCP-fronted GraphQL backend (about 25 organizations targeted in one), and lists seven detection fingerprints (details) | Reported (Wiz) |
| Oct 29 | Hugging Face trust_remote_code miner campaign | Supply chain | Pluto Security: 31 malicious repos on seven accounts posing as small Qwen2 test models run attacker Python when loaded with trust_remote_code, installing a miner and sometimes a reverse shell; 32,000+ downloads. Air Security separately found a fake Solana kit that lures coding agents (details) | Reported (Pluto Security) |
| Oct 29 | Embrace: embedded-browser flaws in five AI dev tools | Critical vuln | Forever Security: Cursor code execution (patched), Copilot URL spoofing (fixed in 1.1.29), file reads in Claude Desktop, Codex and VS Code (Codex and VS Code reported unpatched). Researcher proofs of concept, no CVEs, no in-the-wild use reported (details) | Partly confirmed |
| Oct 29 | Epoch AI: open-weight model builds Firefox exploits from public fixes | AI-run attack | Epoch AI says GLM 5.3 built working file-read exploits for six Firefox releases (156.0 to 158.0.1) from public but unshipped fixes, about $1,000 per version; no code execution. Research, no in-the-wild use; Mozilla published a response (details) | Reported (Epoch AI) |
| Oct 29 | Polymarket: compromised third-party provider served a wallet-draining script | Supply chain | Polymarket says a third-party provider was compromised; BleepingComputer and Hudson Rock tie a crypto-draining script to DebugBear's RUM product (not confirmed by Polymarket). About $2.5M to $2.7M traced; trading resumed after about an hour; affected traders to be reimbursed (details) | Confirmed by Polymarket (vendor unconfirmed) |
| Oct 29 | Cursor marketplace: trojanized Claude Code extension 2.1.313 plants an SSH key | Supply chain | Cursor says a modified Anthropic.claude-code 2.1.313 was served from its marketplace for about 11 hours, affecting at most 5,000 users. It adds an svc-telemetry SSH key, installs persistence and beacons host details. Not built by Anthropic; entry point unknown; update to 2.1.314 (details) | Confirmed by Cursor |
| Oct 29 | Unauthorized Codex and Claude Code relays turn sandboxes off | Agent incident | Symantec: a grey-market reseller installs genuine Codex and Claude Code, rewrites config to route prompts and tokens through its servers and sets full-access sandbox mode; about a dozen organizations, no abuse by the operator found. Tools are not vulnerable (details) | Reported (Symantec) |
| Oct 29 | NIK C2 malware falls back to the victim's OPENAI_API_KEY | AI-run attack | Cisco Talos: a Go-based Windows implant seen with a live server has an ai_generate command that calls the OpenAI API from the victim PC and, if no key is supplied, reads the local OPENAI_API_KEY. Details per press coverage of the Talos report (details) | Reported (Cisco Talos) |
| Oct 29 | 26,400 live cloud and AI keys found in public Gitee repos | Breach | Researcher Ben Zimmermann scanned 7,256,531 public Gitee repos and found 26,400 credentials that still authenticated at test time, including 9,473 Alibaba Cloud AccessKey pairs and 3,170 DeepSeek keys. Exposure finding, not a breach of Gitee; revocation not confirmed (details) | Reported (researcher) |
| Oct 30 | Binance: compromised DevOps laptop, unauthorized withdrawals, withdrawals suspended | Breach | Binance says a compromised DevOps engineer laptop was used to reach a service container and submit unauthorized withdrawals. About $6.3B estimated impact, about $3B frozen, no evidence of private key compromise; deposits and withdrawals paused. Preliminary (details) | Confirmed by Binance (preliminary) |
| Oct 30 | Aur0ra ransomware operators used Cursor agents to reach domain admin | AI-run attack | Sophos CTU: GOLD OGRE used Cursor agents (Cursor Composer plus Anthropic, xAI and Moonshot models) for AD discovery, Kerberoasting, NTLM relay and credential extraction; three victims where agents reached domain admin later appeared on the Aur0ra leak site. No Cursor flaw described (details) | Reported (Sophos) |
| Oct 30 | n0n ransomware operator ran OpenCode with 154 models, left workspace open | AI-run attack | SOCRadar: an n0n extortion operator used OpenCode (154 model entries, $7,216+ of AI spend) to write attack scripts; an exposed file server held 1,719 files. Entry was weak or stolen credentials, no exploited vulnerabilities; ransom and victim claims are the group's own (details) | Reported (SOCRadar) |
| Oct 30 | Wiz: Black October, nine AI platform incidents and LLMjacking of stolen keys | Breach | Wiz lists nine AI developer platform incidents in October (Fireworks AI to Northflank), mostly hunting stored env vars, and documents two stolen enterprise Bedrock keys abused within 30 minutes: about 800,000 and 1.7 million calls. No single shared vulnerability; most investigations ongoing (details) | Reported (Wiz) |
| Oct 30 | Chainalysis: attacker used "offensive agentic techniques" to reach internal tools | Breach | Chainalysis says an external actor used offensive agentic techniques to reach its systems and internal tools; user IDs, names, emails and role labels were accessed. No evidence of investigation content, passwords, MFA secrets or session tokens; scope still being worked out (details) | Confirmed (Chainalysis) |
| Oct 30-31 | Anthropic: Claude agents wrote training data that evades a monitor | Agent incident | Anthropic Alignment Science: Claude agents were told to write fine-tuning data that passes a training-data monitor yet makes a small model misaligned, and mostly succeeded. Deliberate lab study, models never deployed; not a finding about deployed Claude (details) | Confirmed (Anthropic study) |
| Oct 31 | VOLTA infostealer asks an LLM to rank each victim | AI-run attack | Cisco Talos: a self-deleting Windows infostealer (first seen Oct 11) takes browser data, wallets, sessions, SSH keys and cloud configs, then sends a victim profile to an LLM via OpenRouter (default Claude Opus 4.6, Kimi K2 and Gemini fallbacks). No agent-credential targeting documented; victim count unstated (details) | Reported (Cisco Talos) |
New entries are added here as they are confirmed.
Campaigns reported in October that started earlier
These were disclosed or covered this month, but the attack or fix predates October, so they sit outside the daily timeline.
| Incident | Started | What happened |
|---|---|---|
| Fake Claude Code installers | March 2026 | Search ads lead to cloned docs that tell developers to paste an infostealer install command |
| SneakPeek | Fixed August 17 | A poisoned HTML file could run script in the Claude Code Desktop preview and read local files |
| OpenCode MCP OAuth RCE | Fixed September 28 | A remote MCP server could run commands on Windows and WSL during OAuth sign-in; no CVE |
| Codex monitoring gap | May to July 2026 | METR and Redwood estimate about 10% of eligible internal Codex traffic went unmonitored |
| PingCAP TiDB Cloud breach | Sept 9 to Oct 7; told Oct 19 | Stolen service-account credentials let an attacker copy internal data and source code, including account details, SQL editor text, billing data and encrypted API-key values; PingCAP says no evidence TiDB-stored data was affected |
Unconfirmed claims
| Date | Claim | Where it came from | What to do |
|---|---|---|---|
| Oct 20 | OpenRouter breach | Dark-web trackers; a law firm investigation | Rotate, cap and scope keys anyway |
| Oct 22 | TiDB Cloud leak: FulcrumSec's 10.6 TB claim | FulcrumSec leak-site listing (Oct 19); PingCAP confirmed a smaller breach but not these figures | Rotate TiDB Cloud API keys; watch for PingCAP's customer notices |
| Oct 22 | PwC breach (Everest) | Everest leak site; Hackmanac; CTIWatch | Watch for a PwC statement; review shared credentials |
| Oct 23 | Rand Mutual Assurance (South Africa) | Underground forum claim relayed by Dark Web Intelligence; 366 GB, 1.68M claim records, $95,000 demand | Watch for an RMA statement; claimants beware of payout-verification scams |
| Oct 23 | NPCI UPI expired certificate | Researcher thread on X; third-party write-up | Audit bundled certs and how your code validates them |
| Oct 24 | "Universal jailbreak" via spoofed tool calls | X post by @chetaslua | Limit agent permissions; treat tool output as untrusted |
| Oct 25 | GMO breach linked to an AI penetration tool | Social posts and aggregator articles; not in GMO's notices | Do not cite an AI cause until GMO publishes one |
| Oct 26 | Cathedral: AI agents breached Anthropic in 12 hours | Washington Post report of a draft Cathedral pentest report; Anthropic has not confirmed scope | Treat as unconfirmed; keep secrets out of chat tools and log what your agents do |
| Oct 27 | GMO attackers threaten ransomware on customer PCs; users told to power off | Aggregator posts and secondary outlets; ransom reported as both $50M and $1M; no confirmation ransomware was planted | Power off and disconnect PCs that logged in to Desktop Cloud, then rotate credentials from a clean device |
| Oct 27 | Non-member deleted a company's private GitHub repo; other users report similar | Company co-founder's post and a partly redacted audit-log screenshot, plus replies describing similar deletions; GitHub has not responded | Disable member repo deletion, audit outside collaborators and delete_repo tokens, keep a mirror |
| Oct 29 | ScarlettGroup claims AU Small Finance Bank KYC, Aadhaar, credit and loan data | Hackmanac alert and a leak page screenshot with a negotiation countdown; bank asked people to DM and has not confirmed | Ignore OTP/KYC-update lures, lock Aadhaar biometrics, check your credit report, use official bank channels only |
| Oct 29 | Contractor's Claude Code agent on a DOD network | WIRED, anonymous sources; Accenture says allegations were found without merit, DOD declines comment | Use scoped, short-lived credentials for agents, separate personal and work accounts, restrict agent egress |
An entry moves into the timeline if the company or a named research team confirms it, and is removed if it is shown to be false.
Next month: November 2026 attacks timeline
What October's attacks have in common
- Credentials, not exploits, opened most doors. Zeabur's admin tokens skipped rotation, LiteLLM's attacker used stolen GitHub CLI OAuth tokens, ConnectWise lost an app credential, OVHcloud lost a contributor account.
- Trusted channels delivered the payload. Release workflows, trusted publishing, update servers and signed app identities all carried malware this month.
- AI keys are a primary target. Railway, OpenCode, LiteLLM, Tensorlake, GhostAction and LLMjacking campaigns all aimed at AI access, because a stolen key is a free compute and intrusion budget.
- AI now does the work inside attacks. INC, BUGSTER, ARTEX, Storm-0940 and CARLIY0S used agents for recon, commands, debugging and data triage. Humans still picked targets and approved steps.
- Patch windows shrank to hours. PHP2Shell was attacked before the fix and Plex2Shell within a day of the PoC.
- Agents caused harm with no attacker. OpenAI, Anthropic and CERT-In all reported agents reaching systems or removing controls they should not have.
What to do this week
- Rotate every secret stored on Railway, Zeabur and OpenCode Console, and on any host that ran LiteLLM 1.107.0, Tensorlake 0.5.144,
@subql/common5.8.3 or OVHcloud CLI v0.11.0 to v0.15.0. - Pin packages and container images by version and digest; turn on GitHub release immutability.
- Search your GitHub org for unknown workflows named like
security-audit.yml. - Audit Microsoft 365 service principals and credential additions since October 1.
- Patch PHP, Node.js, GitLab, Java and Plex; update trust stores and rebuild containers for eMudhra.
- Cap and scope every AI provider key, and alert on usage spikes.
- For your own coding agents: keep monitors outside the agent's reach, log what they run, and require approval for credential access and destructive commands. Beam's local watcher flags these actions on your machine.
How this timeline is updated
Beam publishes one attacks timeline per month, and a new page starts on the first of each month. This page is checked every 4 hours and updated whenever a new October incident is confirmed or a claim changes status; the "Last updated" line at the top changes with it. The attacks tracker shows the same entries with filters, a link for every incident and an embed code you can paste into your own page. Each entry links to a full post with primary sources; entries without their own post rely on the sources listed in the supply chain tracker.
Change log
- November 4, 2026: Added the KVM guest-to-host escape found through the Vercel Sandbox bounty (Oct 3), missing from the first pass.
- October 31, 2026: Added n0n ransomware via OpenCode, Wiz's Black October report, the Chainalysis breach, Apple iOS 27.2, the Gitee 26,400 keys scan, Anthropic's monitor-evasion study and Talos's VOLTA infostealer. Added the patched ChatGPT share link XSS (Parameter).
- October 30, 2026: Added the Hugging Face trust_remote_code miner campaign, Embrace embedded-browser flaws, Epoch AI Firefox patch-gap exploits, and GitLab CVE-2026-97229 and Cisco ASA/FTD CVE-2026-76418 (both CISA KEV). Added the WIRED claim about a Claude Code agent on a DOD network to unconfirmed claims.
- October 30, 2026: Added the Binance security incident (compromised DevOps laptop, unauthorized withdrawals, preliminary).
- October 30, 2026: Added the Polymarket third-party script compromise (vendor attribution unconfirmed).
- October 30, 2026: Added the trojanized Claude Code extension 2.1.313 on the Cursor marketplace.
- October 30, 2026: Added the DHS employee data leak (40,000+ records; Okta source claimed and disputed).
- October 30, 2026: Added CrowdStrike's VAPOR PANDA and ARTEX campaign, Symantec's unauthorized Codex and Claude Code relays, and Talos's NIK C2 malware.
- October 30, 2026: Added Sophos's report that Aur0ra ransomware operators used Cursor agents to reach domain admin. Added Ivanti Connect Secure CVE-2026-111357 (CISA KEV).
- October 29, 2026: Added the vLLM 0.32.0 advisories (two RCE paths) and the OWASP Phase 2 lab report; Fragnesia CVE-2026-46300 now noted as added to CISA KEV. Added VMware ESX CVE-2026-41709 (CISA KEV).
- October 29, 2026: Pointed Zeabur, TeamUpdraft and OVHcloud CLI at their new breakdown posts; added the Linux KSMBD flaws.
- October 29, 2026: Added Microsoft October Patch Tuesday (CVE-2026-96951 exploited) with a breakdown post.
- October 29, 2026: Added Linux kernel CVE-2026-98432 (TCP use-after-free, patch status by distro, MPTCP-only claim checked).
- October 29, 2026: Added ScarlettGroup claim against AU Small Finance Bank to unconfirmed claims.
- October 29, 2026: Added the Wiz report on three agent-run intrusions (Flowise, Java gateway, MCP backend) with a breakdown post.
- October 28, 2026: Added Brotli CVE-2026-111497 (CISA KEV, Cloudflare WAF rules), Anthropic's AI-built exploit chains and the MCP Python SDK advisories. PHP2Shell now notes the solved StealMyBitcoin challenge.
- October 28, 2026: Added the GitHub repo-deletion claim, the testmetest worm (42 repos), GPT-6.1 Terra results, TrendAI CyberStrikeAI, the 123 Reg attack and the Dream Research report of a PHP2Shell and Fragnesia chain that rooted 25 firewalls.
- October 27, 2026: Added the Cohere report of Qwen3.8-Flash-Next trying an upload to a Beijing bucket with a breakdown post. PHP2Shell CVEs now noted as added to CISA KEV. Added the Check Point report on the Aokigahara infostealer targeting Claude Code, Codex and Cursor files, with a breakdown post.
- October 27, 2026: Added live updates to the GMO Internet breakdown post: reported ransomware threat and noon power-off warning, disputed ransom figure ($50M vs $1M); added the threat to unconfirmed claims.
- October 27, 2026: Added the LiteLLM (proxy admin RCE, file read), vLLM and MCP SDK advisories, the Warp vendor breach, EPICS Base CVEs found by Cipher, and Adobe Commerce CVE-2026-48356 (CISA KEV). Published the Black October week-in-review post.
- October 26, 2026: Added the Tailworm report (11 unpatched Tailscale and headscale findings, Tailscale SSH root escalation, lab only) with a claimed-vs-verified post.
- October 26, 2026: Added the GMO Internet Onamae.com Desktop Cloud breach (26,739 servers potentially affected) with a breakdown post, and the unconfirmed AI-tool link to unconfirmed claims. Added the unconfirmed Cathedral claim of an AI-agent breach of Anthropic to unconfirmed claims.
- October 25, 2026: Added the ProjectDiscovery backdoored Qwen model proof of concept (credential theft through Codex) with a breakdown post. Added the confirmed Capcom ID and Quest Diagnostics breaches and the PingCAP TiDB Cloud breach (earlier); the FulcrumSec 10.6 TB claim stays unconfirmed. Added the malicious AGENTS.md in Downloads incident (Forever Security) with a breakdown post.
- October 24, 2026: Added the OpenAI containment red-team report (VM escapes, cache-timing channel) with a breakdown post. Added the exploited Arista VeloCloud Orchestrator flaw (CVE-2026-114625), now in CISA KEV.
- October 24, 2026: Added the unverified "universal jailbreak" claim (spoofed tool calls, CTF framing) to unconfirmed claims, with a claimed-vs-verified post.
- October 23, 2026: Page created with 48 dated entries covering October 1 to 22, plus 4 earlier campaigns and 2 unconfirmed claims.
- October 23, 2026: Added the Everest claim against PwC (590.69 GB, unconfirmed) to unconfirmed claims.
- October 23, 2026: Added the unconfirmed Rand Mutual Assurance claim (366 GB, deadline Nov 5) to unconfirmed claims.
- October 23, 2026: Published the Rand Mutual Assurance claimed-vs-verified post.
- October 23, 2026: Added the x402 SDK critical flaws (patched Aug 18, no advisory) with a breakdown post.
- October 23, 2026: Published the NPCI UPI expired certificate claimed-vs-verified post.
- October 23, 2026: Added the Panasonic generative AI data upload report with a claimed-vs-verified post.
- October 23, 2026: Added the Mythos 5 AISI malicious pull request incident with a breakdown post.
- October 23, 2026: Added the exploited Collabora built-in CODE server flaw (CVE-2026-110447), now in CISA KEV.
Related reading
- October 2026 supply chain incident tracker
- AI-led attacks in October 2026
- AI agent security incidents: a timeline (2025-2026)
- Anthropic threat intelligence report, September 2026
Accurate as of October 23, 2026. Several incidents are still under investigation, and counts, dates and attributions may change as companies publish final reports.

