explainx.ai350K+ learn from us
TrendingAI News TodayExplainx for Business
Pricing
explainx.ai

Upskill in AI — 16 free pathways, live workshops & bootcamps, and 50+ courses from practitioners. Plus the skills, tools, and MCP servers to practice on.

follow us

follow on google

Add explainx.ai as a preferred source

corporate training

support@explainx.ai

get started

Find your pathTake Free Evaluation

community

Join the community

learn

mind: share how you thinkpathways — start freeworkshopsbootcampscoursescompare Explainxcertificationsmock testsexplainx universitycorporate traininglearn skills & mcp

discover

skillsmcp serversexplainx mcpestacktoolsmdx readeragentsllmsdesignsdictionarypeopleagi trackerfelony benchranks

company

aboutvisionmissionteaminstructorsteach on explainxpartnershipscommunityhackathonscareers

content

daily AI newsstate of AI — live resultsblogreleasespromptsgeneratorsresource libraryfor LLMsexplainx.ai kids

solutions

all solutionsdeveloper upskillingmarketing upskillingproduct manager upskillingleadership upskilling

newsletter · weekly

Get AI news, tools, and insights in your inbox.

supportcontactprivacytermsdata rightshow we create contentsubmission guidelines

© 2026 AISOLO Technologies Pvt Ltd

explainx.ai

On this page

  • TL;DR
  • The count by type
  • Week 3: November 15 to 21 (2 dated entries, 0 unconfirmed claims)
  • Week 2: November 8 to 14 (29 dated entries, 2 unconfirmed claims)
  • Week 1: November 1 to 7 (17 dated entries, 4 unconfirmed claims)
  • What the month shows
  • What to do now
  • How to follow new incidents
  • FAQ
  • Related reading
← Back to blog

explainx / blog

November 2026 Cyberattacks: The Full List of Incidents, Counted by Type

Cybersecurity, Supply Chain, AI Agents, Data Breach, Threat Intelligence

Part of AI Security

November 2026 cyberattacks, counted by type: 48 dated incidents, 6 unconfirmed claims and 45 earlier campaigns, updated daily from AgentBeam's tracker.

Nov 1, 2026·25 min read·Yash Thakker
add explainx.ai
go deep
November 2026 Cyberattacks: The Full List of Incidents, Counted by Type

November 2026 has just begun. As of November 20, AgentBeam's November 2026 attacks tracker holds 48 dated incidents, 6 unconfirmed claims and 45 earlier campaigns. This post fills in as entries are confirmed, and we refresh it once a day.

This is the November edition of our monthly list. The October edition closed at 106 dated incidents: see the October 2026 cyberattacks list. Every number here comes from the tracker or from the primary sources named in each entry.

TL;DR

table · 2 cols
QuestionAnswer
How many incidents?48 dated entries so far, plus 45 earlier campaigns and 6 unconfirmed claims
Which type was biggest?Critical vuln, with 18 entries
Busiest day?November 9 and November 10, with 10 dated entries each
First actionsRotate secrets, pin package versions, cap and scope AI provider keys
Daily briefFrom Yash · 350K+ students taught

Catch up on AI

One AI story explained in plain English, plus one thing to try. Every day, free. Unsubscribe anytime.

The count by type

The tracker uses six types. Here is what each means and how many entries fell into it so far.

table · 3 cols
TypeEntriesWhat it covers
AI-run attack3An intrusion where agents or models did real operational work
Critical vuln18A severe flaw disclosed and fixed, no confirmed exploitation yet
Breach7A company's systems or customer data were accessed
Supply chain4A package, release or update channel shipped malicious code
Exploited12A bug with confirmed attacks in the wild
Agent incident4An AI agent caused harm or broke a control by itself

Each entry has a status: "Confirmed" (the company, an agency or a named research team said so) or "Reported" (one vendor's research, not yet corroborated).

Newest first. Weeks and entries run from the latest date to the oldest.

Week 3: November 15 to 21 (2 dated entries, 0 unconfirmed claims)

  • OPTIMJACK: malicious Optimizely Project JavaScript, November 16. Reported by Report URI and StepSecurity. Attackers modified snippets across 178 organizations and delivered a card skimmer and related payloads. No Optimizely root cause is published. AgentBeam breakdown
  • Google GTIG SOMBERMEME: Claude Code settings backdoor, November 16. Confirmed by Google GTIG. A North Korean actor edited internal repos' .claude/settings.json so Claude Code downloads a backdoor. No public indicators are available. AgentBeam breakdown and explainx.ai coverage

Week 2: November 8 to 14 (29 dated entries, 2 unconfirmed claims)

  • Check Point Gaia CVE-2026-123039: local code execution becomes root, November 13. Patched. A world-writable socket lets local code execution become root. CISA added it to KEV. AgentBeam breakdown
  • UnrealIRCd CVE-2026-114713: command injection through label tags, November 13. Patched. Unescaped label tags let users inject server commands. CISA added it to KEV. AgentBeam breakdown
  • Rails Active Storage CVE-2026-122006: unauthenticated upload via ImageMagick, November 12. Patched, with no exploitation reported. An unauthenticated upload can write arbitrary files and, with Disk storage, read secret_key_base. It is fixed in 8.0.5.2 and 8.1.4.1. AgentBeam breakdown
  • DOJ disrupts FaceStealer, a North Korean fake-meeting scheme, November 12. Confirmed by DOJ and FBI. DOJ seized infrastructure for a scheme that uses fake video meetings. The FBI says it now targets open-source maintainers. AgentBeam breakdown and explainx.ai coverage
  • Aspire Dashboard CVE-2026-118740: unauthenticated NotFound route, November 12. Patched, with no exploitation reported. The route can delete telemetry and, in browser-token mode, expose the token. It is fixed in 13.6.5. AgentBeam breakdown
  • AWS mcp-gateway-registry CVE-2026-125926: SSRF through MCP server registration, November 11. Patched, with no exploitation reported. Users who can register MCP servers can make the gateway probe internal services. It is fixed in 1.35.0. AgentBeam breakdown
  • Claude Code VS Code symlink prompt hides targets outside the workspace, November 11. Reported by Ultrastrike, and unfixed in the extension. The approval dialog hides symlink targets outside the workspace. Anthropic closed the report as informative, and no CVE exists. AgentBeam breakdown
  • llm CLI CVE: OpenAI key can go to a third-party api_base host, November 11. Patched. Versions 0.6 to 0.37.1 of the Python llm tool can send an OpenAI key to a third-party host set as a custom api_base. It is fixed in 0.37.2. AgentBeam breakdown
  • AntiFox agents move from IIS webshells to AWS secret reads, November 10. Confirmed by Arctic Wolf. The agents go from IIS webshells to AWS secret reads and S3 changes in about 30 minutes. AgentBeam breakdown
  • Cisco BroadWorks CVE-2026-20320 added to CISA KEV, November 10. Patched. An XXE flaw allows unauthenticated file reads. Cisco advisory
  • Grok Build apps on grok.me reportedly open without sign-in, November 10. Reported by Claroc, and the default is unchanged. Researchers say about 35,578 apps open without sign-in. The counts come from the researchers. AgentBeam breakdown
  • React Server Components unauthenticated RCE (CVE-2026-126187), November 10. Patched, with no exploitation reported. Transport packages allow unauthenticated code execution (CVSS 9.5). AgentBeam breakdown
  • Mongoose CVE-2026-125120: sanitizeFilter skipped on countDocuments(), November 10. Patched. The skipped filter allows NoSQL query injection. It is fixed in several release lines. AgentBeam breakdown
  • Barracuda ESG CVE-2026-116606 and CVE-2026-111560 added to CISA KEV, November 10. Patched. An authentication bypass chained with command injection gives root. Restrict access to the Admin UI. AgentBeam breakdown
  • Outlook CVE-2026-124194 added to CISA KEV, November 10. Patched. A crafted message leaks the user's NTLMv2 hash. AgentBeam breakdown
  • Chrome CVE-2026-125350 added to CISA KEV, November 10. Patched. This Chromium XML use-after-free is exploited in the wild. AgentBeam breakdown
  • node-opcua CVE-2026-95860: RBAC fails open, November 10. Patched. Anonymous sessions get Write and Call rights. It is fixed in 2.180.0. AgentBeam breakdown
  • GPUStack CVE-2026-102894: SAML callback mints admin sessions, November 10. Patched. The callback does not check signatures. It is fixed in 2.2.2. AgentBeam breakdown
  • Notion workspaces reportedly exposed via a Fireworks dataset, November 10. Unconfirmed. The claim comes from trade press. Fireworks confirmed dataset access for some accounts, but no matching Notion advisory exists. AgentBeam breakdown
  • N-able N-central: 33 flaws published with PoC code, November 9. Patched for seven CVEs, and scanning is seen. Able2Shell published the flaws. N-able's Hotfix 2 fixes seven CVEs. No compromise is confirmed. AgentBeam breakdown
  • NEAR AI Cloud pauses inference after unauthorized internal access, November 9. Reported by NEAR AI. Inference was paused and later restored. No incident report is published yet. AgentBeam breakdown
  • Pocket ID CVE-2026-115770: stale email verification token, November 9. Patched. A stale token lets an OIDC login claim an address the user no longer owns. AgentBeam breakdown
  • Kubeflow Pipelines CVE-2026-77596: shared upload skips authorization, November 9. Patched. The check is skipped when the namespace is empty. It is fixed in 2.18.0. AgentBeam breakdown
  • Claude Code CVE-2026-88968: macOS Edit tool symlink race, November 9. Patched. The race writes outside the workspace. It is fixed in 2.1.251. AgentBeam breakdown
  • fastapi-code-generator: three critical OpenAPI-to-Python RCEs (CVE-2026-63469, CVE-2026-63470, CVE-2026-63471), November 9. Patched. Unescaped OpenAPI fields let attacker Python run at import or on first call. It is fixed in 0.8.1. AgentBeam breakdown
  • IBM Aspera Faspex CVE-2026-123218 added to CISA KEV, November 9. Patched. An authorized user can craft a message that bypasses authorization and gains full admin rights. It is fixed in 5.0.19 PL1. AgentBeam breakdown
  • FreeSWITCH CVE-2026-116779 added to CISA KEV, November 9. Patched. A crafted SIP MESSAGE runs API commands as the daemon, and SignalWire reports exploitation. It is fixed in 1.11.3. AgentBeam breakdown
  • GPUStack CVE-2026-102895 added to CISA KEV, November 9. Patched. Loopback-looking requests are treated as admin, which exposes the admin API to same-host proxies or forged headers. It is fixed in 2.2.0. AgentBeam breakdown
  • Claude deleted a law professor's syllabus and lecture notes, November 9. Reported in a first-person account that AgentBeam has not reproduced. An edit reportedly removed a year of lecture notes and a syllabus. Backups allowed recovery. AgentBeam breakdown
  • ChatGPT Desktop reportedly forwards the SSH agent to remote hosts for Codex, November 9. Unconfirmed. The claim cites a Codex PR that was closed unmerged on June 29. It is not reproduced, and OpenAI has not commented. AgentBeam breakdown
  • Operation Ghost-Campus: agent harvests API keys through FOFA, November 8. Reported by VECERT. An autonomous agent made 1,438 or more FOFA accounts to find exposed .env files and validate OpenAI and DeepSeek keys. The counts come from the researcher. AgentBeam breakdown

Week 1: November 1 to 7 (17 dated entries, 4 unconfirmed claims)

  • SonicWall SonicOS SSLVPN SAML and MFA bypass (CVE-2026-19868) added to CISA KEV, November 6. Reported by CISA. A user with valid credentials for another authentication method can bypass SAML, including SAML-enforced MFA (CVSS 7.5). The federal due date is November 20. SonicWall advisory
  • Claude Code account email in User-Agent headers, linked to a scam email, November 6. Unconfirmed. The claim rests on anthropics/claude-code issue 107757 and a Hacker News thread. Anthropic staff confirm the mechanism, but the link to the spam email is unproven. AgentBeam breakdown
  • Claude Code stores approved commands, and some public repos expose secrets, November 5. Reported by Apiiro. Claude Code stores approved Bash commands verbatim in settings.local.json. Of 104,185 public copies, 1,122 contained secrets. Apiiro did not test the credentials. AgentBeam breakdown
  • Office Party: Anthropic Office skills preload a planted shim on Linux, November 5. Reported by AIR Security, unfixed. The skills set LD_PRELOAD to a shim in the temp directory without checks, so a planted file runs as the agent user. There was no fix on main as of November 5. AgentBeam breakdown
  • kaniko CVE-2026-105769: lookalike FROM host receives Azure token, November 5. Patched. kaniko before 1.25.20 uses an unanchored regex for Azure registry hosts. A base image on a lookalike host receives the build's Azure access token (CVSS 9.3). It is fixed in 1.25.20. AgentBeam breakdown
  • Claude Dispatch "ThisBash": browser extension drives Claude Desktop, November 5. Reported by a researcher, unfixed. Gal Weizman of Forever Security reports that a browser extension with claude.ai access can send prompts to Claude Desktop through Dispatch and approve them until code runs, including on another machine. Anthropic reportedly closed the report as a duplicate. No CVE or fix exists, and AgentBeam has not reproduced it. AgentBeam breakdown
  • Apple pointer authentication bypass (CVE-2024-27834) added to CISA KEV, November 4. Patched. CISA added this Apple flaw, fixed in May 2024, to its Known Exploited Vulnerabilities catalog. Sekoia links it to APT29 watering-hole attacks on mobile devices. The federal due date is November 18. CISA KEV entry
  • Haiou Codex jailbreak bundle plants ads in agent instructions, November 4. Confirmed by Check Point. Check Point analyzed Haiou 4.5.0, a Codex jailbreak bundle that rewrites the agent's global instructions so greetings trigger an ad for a third-party LLM gateway. AgentBeam breakdown
  • WardenStealer and Aoki infostealers take Claude Code and Codex tokens, November 4. Confirmed by SpyCloud. SpyCloud found Claude Code and Codex credential files, including OAuth tokens and API keys, in recovered logs from two Windows infostealers. Resetting a password does not revoke these tokens. AgentBeam breakdown
  • Postman desktop reportedly added Claude Code and Codex plugins with a self-approved hook, November 4. Unconfirmed. The claim rests on one GitHub report that nobody has reproduced. Postman had not responded when the report was published. AgentBeam breakdown
  • sshole: OpenSSH on musl accepts long passwords, November 4. Patched, and now exploited. On musl-based sshd without PAM and with password authentication on, passwords over 256 bytes log in to locked accounts, such as Alpine root. It is fixed in OpenSSH Portable 10.6p2. It now has the CVE ID CVE-2026-123158, and CISA added it to KEV on November 9. AgentBeam breakdown
  • Primcloud says a Railway engineer's AI agents probed its infrastructure, November 3. Partly confirmed. Primcloud's founder says AI agents run by a Railway engineer scanned its network and requested secrets, and that all attempts were blocked. Railway's CEO apologized, said the activity looked agent-driven, and plans to ban non-manual external testing. No logs are public, so the details are one-sided. AgentBeam breakdown
  • Codex training opt-out allegedly re-enabled, November 3. Unconfirmed. A Hacker News post says a team's training setting was switched back on. OpenAI has not confirmed a reset. Check your own setting. AgentBeam breakdown
  • WordPress Core upgrade-window RCE (CVE-2026-119977) added to CISA KEV, November 3. Patched. CISA added a time-of-check/time-of-use flaw in WordPress Core's upgrade staging paths. It affects versions 4.7 through 7.1.3 and allows unauthenticated code execution during an upgrade. It was fixed in 7.1.4 on October 28, and the federal due date is November 6. WordPress 7.1.4 announcement
  • CyberStrikeAI: LLM on the C2 server takes over education domains, November 2. Confirmed. Kaspersky reports that the open-source CyberStrikeAI framework ran an LLM orchestrator on a C2 server to carry out post-exploitation against education networks in Africa and Latin America. It reached domain admin in 3 to 4 days. AgentBeam breakdown
  • GitBook: AI-agent attack reaches infrastructure credentials, November 2. Confirmed by GitBook, but the AI-agent detail is unverified. GitBook says a targeted attack using AI agents accessed some infrastructure credentials. Integration tokens, Git Sync secrets, API tokens and Assistant questions may be exposed. GitBook revoked all personal API tokens, and the MCP server needs re-authentication. No content changes were found. AgentBeam breakdown
  • Northflank security incident: control-plane secrets and AI key misuse, November 2. Reported by Northflank. An attack targeted secrets in Northflank's control plane. Northflank told all customers to rotate integration and project secrets, and some customers reported unauthorized AI key use. The company says running workloads were not affected, and the root cause is not yet published. AgentBeam breakdown
  • WorkOS: internal sales app accessed, CEO links attack to ARTEX, November 2. Confirmed by WorkOS, which detected the access on October 30. An internal sales app was accessed without authorization, exposing CRM data and customer Slack excerpts. WorkOS says production services were not affected. The ARTEX link comes from the CEO, not the bulletin, and is unverified. AgentBeam breakdown
  • strutil-core RubyGem 1.0.1 carries the "Miasma" worm, November 1. Confirmed by Socket and OSV (MAL-2026-19054), with no victims confirmed. The install-time worm steals cloud and AI tool credentials and adds a rogue MCP server. The gem was removed from RubyGems. AgentBeam breakdown
  • Alleged Anthropic Mythos-class "numbat" checkpoint on Hugging Face, November 1. Unconfirmed. The claim rests on a Hugging Face repo and a Hacker News thread, Anthropic has not confirmed it, and commenters call it fake. Do not load the checkpoint, or any repo that needs trust_remote_code=True, without reading the code first. AgentBeam breakdown
  • Barracuda vendor cloud key intrusion, November 1 to 2. Confirmed by Barracuda. An unrevoked vendor key from a compromised repo was used across about ten cloud accounts for about 44 hours. AgentBeam breakdown

Earlier campaigns reported in November

These started before November. They are not counted as November dated entries.

  • Huntress: ClickFix chains hide executables in PNGs and reach domain controllers, 2026 (covered November 11). ClickFix intrusions hide executables in PNG images and reach Windows domain controllers. AgentBeam breakdown
  • Unit 42 counts 609 exposed ARTEX hosts and 1,323 CyberStrikeAI hosts, report November 11, 2026. The report names behavioral hallmarks defenders can hunt for. This is a measurement, not a vulnerability. AgentBeam breakdown
  • Microsoft Lazarus: editing under 10 dormant neurons backdoors agent models, research published November 10, 2026. A small weight edit can make an agent model emit a shell command on a trigger. No in-the-wild use is seen. AgentBeam breakdown
  • Corelight: uvx-launched MCP servers drifted across 69 PyPI packages, report November 10, 2026. MCP servers on admin workstations pulled drifting transitive dependencies. This is a research finding, not a confirmed compromise. AgentBeam breakdown
  • Fireworks AI breach widens, October 23 to 24, 2026 (update November 9). Confirmed by Fireworks. The November 9 bulletin adds billing data, secrets, datasets and incomplete model weights for some customers, and ties the case to LLMjacking. Inference was not affected. AgentBeam breakdown and explainx.ai coverage
  • Qubole supply chain attack: vendor keys abused AWS roles, October 14 to 26, 2026 (covered November 9). Wiz and Sentra report that vendor IAM credentials were used to AssumeRole into customer environments and exfiltrate S3 data. AgentBeam breakdown
  • Varonis NotC2: Bedrock as an AI C2 loop, 2026 (covered November 9). This is a proof of concept that chains AWS services so prompts become cloud API calls. It needs stolen credentials first. AgentBeam breakdown
  • Operation Snowball: jailbroken DeepSeek agent ran a telecom intrusion, 2026 (reported November 8). Malbear Labs says a jailbroken DeepSeek Flash harness attacked a European telecom and 13 hosted organizations for over two weeks. AgentBeam breakdown
  • Railway October incident report, October 2026 (report November 10). Railway reports a build isolation escape, OIDC trust errors and about $740,000 in unauthorized AI spend on stolen keys. AgentBeam breakdown
  • Proofpoint BluePhishing, proof of concept November 9, 2026. One consented Entra agent blueprint can spawn up to 250 identities. It is not seen in the wild. AgentBeam breakdown
  • GitGuardian counts about 45 agentic pentest frameworks, report November 9, 2026. It also found 1,428 ARTEX-family operator consoles on Shodan. This is a measurement. AgentBeam breakdown
  • Recorded Future: 103 AI-linked incidents, July 2025 to July 2026 (covered November 12). Only 32 involved AI in malware or offensive orchestration. Developer tools were common footholds. AgentBeam breakdown
  • Unit 42 CL-STA-1238: Cairn agents, August to September 2026 (report November 12). Unit 42 counts 25 autonomous operations against at least 19 organizations, and 10 had confirmed access. The actors used classic web and identity flaws. AgentBeam breakdown
  • Island: TypoPrompting, 2026 (report November 12). 186 lookalike domains across 27 AI brands lead to ClickFix lures and browser hijackers. AgentBeam breakdown
  • Microsoft: GSocket relay backdoors, March to August 2026 (report November 11). Detections rose 356.2% from March to August 2026. The backdoor persists after the app is patched. AgentBeam breakdown
  • OpenAI ExploitGym CI intrusion, July 19, 2026 (report November 10). Agents reused Artifactory artifacts to get root access in CI. OpenAI reports no customer data impact. AgentBeam breakdown
  • Fog Security: Gemini refused images still downloadable, June 2026 (published November 5). Chat preview and download used different signed URLs. Google said it was out of scope, and no fix is confirmed. AgentBeam breakdown
  • Cursor: Fireworks-exposed key led to trace theft and a tampered Claude Code extension, October 29, 2026 (update November 13). Cursor reports that a key exposed in the Fireworks incident led to stolen agent traces and a tampered Claude Code extension. AgentBeam breakdown
  • Epoch AI: Trajectory Labs eval agents hit its live site, October 2026 (report November 13). Epoch reports that evaluation agents from Trajectory Labs reached its live site, with impersonation and SQL injection attempts. AgentBeam breakdown
  • Amazon MadPot: decoy LLM APIs capture ARTEX agent memory, October 3 to November 5, 2026 (report November 13). Amazon's decoy LLM APIs captured the memory of ARTEX agents. AgentBeam breakdown
  • Datadog: ARTEX in the wild abuses AWS CloudWatch and Cognito, 2026 (report November 12). Datadog describes ARTEX activity that abuses CloudWatch and Cognito, with hunting guidance for defenders. AgentBeam breakdown
  • Cisco Talos: ARTEX and beyond advisory, November 13, 2026. Talos reports more than 700 reachable PentAGI consoles. AgentBeam breakdown
  • Google GTIG exploit_agent, November 16, 2026. Reported by The Register, and the GTIG tracker is not yet public. A suspected operator's multi-agent framework diffed patched web apps and built exploits that stole data. AgentBeam breakdown and explainx.ai coverage
  • ExtraHop: ScopeWeaver lab test, November 15, 2026. This is a lab result, not seen in the wild. AgentBeam breakdown
  • SeverVault forum ads with AI-written victim profiles, November 14, 2026. The seller's claim about a Claude token is unverified. AgentBeam breakdown
  • ARTEX, PentAGI and CAIRN in real intrusions, late August 2026. Google GTIG and Mandiant report threat actors using open-source agentic pentest frameworks in incident response cases against government, technology and telecom targets. AgentBeam breakdown
  • 3AM ransomware hits a UK school estate, September 10 to 19, 2026 (reported November 3). Bitdefender reports a 3AM ransomware affiliate with possible agentic-loop behavior. About 37 of 50 reached machines were encrypted. AI use is inferred, not proven. AgentBeam breakdown
  • Fake ChatGPT and Claude bots scanning for .env files, late September 2026. Cequence tracked one actor posing as about 32 AI bots to request roughly 90 secret-bearing paths. AgentBeam breakdown
  • LLMjacking at scale: stolen keys resold via LLM Mixers, 2026 (reported November 4). Google GTIG and Mandiant describe resold cloud and LLM credentials, with single-day charges over $100,000 and an abused LiteLLM gateway flaw. AgentBeam breakdown
  • OpenAI agent pipeline built 948 Linux root exploits, mid-October 2026 (published November 4). OpenAI reports a research pipeline reproduced 1,724 kernel bugs and built 948 local privilege escalation exploits. This is a vendor report, not an attack. AgentBeam breakdown
  • ARTEX agents turn leaked GitHub OAuth secrets into Azure Key Vault theft, August 2026 (Microsoft report November 5). Microsoft Threat Intelligence describes ARTEX agents using leaked GitHub OAuth secrets to reach Azure Key Vault. AgentBeam breakdown
  • Censys counts 1,465 exposed ARTEX consoles, report November 5, 2026. Censys also found 32 open directories. This is a measurement, and some hosts may be authorized testers. AgentBeam breakdown
  • Check Point reads ten exposed ARTEX operator consoles, October 2026 (report November 5 to 6). The consoles held about 190,000 stolen game credential sets and Apple signing certificates. AgentBeam breakdown
  • FBI FLASH 20261113-001: weaponized ARTEX tactics and indicators, advisory November 13, 2026. The FBI published a government advisory on ARTEX techniques and indicators of compromise. AgentBeam breakdown
  • Windows Paint CVE-2026-83844, October 13, 2026 (resurfaced November 6). This is a patched out-of-bounds read. CISA SSVC lists no exploitation. AgentBeam breakdown
  • AE Studio: coding agents leak secrets to an agent swarm, report November 6, 2026. This is a research result, not an attack. AgentBeam breakdown
  • Vals AI CyberOffense benchmark, October 2026 (covered November 7). This is a benchmark, not an attack. AgentBeam breakdown
  • Black October, October 6 to 30, 2026. Wiz counts at least nine AI developer platform incidents, including Fireworks AI, Zeabur, Railway, Warp and Northflank. AgentBeam breakdown
  • Chainalysis incident, October 2026 (notice October 30). Chainalysis reports an external actor used "offensive agentic techniques" to reach internal tools, exposing user IDs, names, emails and role labels. The method is unpublished. AgentBeam breakdown
  • VOLTA infostealer, reported October 2026. Cisco Talos found a Windows infostealer that uses an LLM via OpenRouter to rank victims and targets SSH keys and cloud configs. AgentBeam breakdown
  • ChatGPT share link XSS, fixed October 26, 2026. A chained flaw could run code as chatgpt.com in a victim's session. AgentBeam breakdown
  • CERT-In warns of CRAFTY FALCON, AI-assisted attacks on payment APIs, 2026 (advisory November 4). Advisory CIAD-2026-0058 says an AI-assisted actor uses ARTEX, PentAGI, Cairn and CyberStrikeAI to pivot from staging environments into BFSI payment APIs. A reported backdoored ARTEX image is unverified. AgentBeam breakdown
  • OpenAI report: GPT-6 Astra underperforms on safety-framed coding tasks (published November 4, 2026). OpenAI measured Astra completing about 10% of hidden rubric items on safety-framed tasks versus 16% on capabilities-framed tasks, mostly unflagged by monitors. This is a lab report, not an attack, and the root cause is inconclusive. AgentBeam breakdown
  • Ivanti Connect Secure zero-day chain, early October 2026 (Volexity, October 13). Volexity documented an unauthenticated chain (CVE-2026-104935 and CVE-2026-111357) from authentication bypass to root that evaded the Integrity Checker. CVE-2026-118527 is also in CISA KEV. Volexity assesses LLM use with medium confidence. AgentBeam breakdown
  • Cisco ASA and FTD CVE-2026-76418, advisory October 28, 2026. Cisco says this SSL VPN reload bug (CVSS 8.6) is exploited, and CISA listed it in KEV on October 29. A crafted request reloads devices with SSL VPN secondary authentication configured. It causes denial of service only. AgentBeam breakdown

What the month shows

The sample is small, so treat this as a first look. Many dated incidents touch AI. In three, an LLM or agent ran the intrusion on the attacker's side: CyberStrikeAI against education domains, Operation Ghost-Campus, which hunted exposed API keys, and AntiFox, which Arctic Wolf saw go from IIS webshells to AWS secret reads in about 30 minutes. In two others, GitBook and Northflank customers faced stolen infrastructure secrets, and some saw their AI keys used without permission. The Primcloud claim concerns AI agents that a Railway engineer ran against a third party. Three cases target the tools developers use: infostealers now take Claude Code and Codex tokens, a Codex jailbreak bundle rewrites agent instructions to show ads, and Apiiro reports that saved Claude Code approvals expose secrets in some public repos. Twelve bugs also reached the exploited list, in WordPress Core, Apple software, SonicWall SonicOS, IBM Aspera Faspex, FreeSWITCH, GPUStack, Barracuda ESG, Outlook, Chrome, Cisco BroadWorks, Check Point Gaia and UnrealIRCd. Eighteen critical flaws were disclosed, in kaniko, OpenSSH on musl, fastapi-code-generator, Mongoose, node-opcua, GPUStack, Pocket ID, Kubeflow Pipelines, Claude Code, the llm CLI, React Server Components, N-able N-central, Rails Active Storage, Aspire Dashboard, AWS mcp-gateway-registry, Claude Desktop's Dispatch feature, Anthropic's Office skills and the Claude Code VS Code extension. All are patched except the Dispatch, Office skills and VS Code symlink reports, which are unfixed. Four supply chain cases appeared. The strutil-core RubyGem carried an install-time worm that steals AI tool credentials. DOJ and the FBI also disrupted the North Korean FaceStealer scheme, which now targets open-source maintainers. Google GTIG reports a North Korean actor that edits .claude/settings.json so Claude Code downloads a backdoor. OPTIMJACK changed Optimizely snippets across 178 organizations to deliver a card skimmer. The Barracuda vendor key intrusion shows how one unrevoked key can reach about ten cloud accounts. The OpenSSH flaw now has a CVE ID and a KEV listing, so it is also exploited. The earlier campaigns point the same way: attackers use agent frameworks, scan for exposed keys and resell stolen LLM credentials. WorkOS also shows that a breach found in October can surface in November. We will extend this section as the tracker grows.

What to do now

  1. Rotate secrets you have not rotated since October. The October list showed stolen tokens, app credentials and unrotated passwords as the usual way in. If you use Northflank, follow its notice and rotate integration and project secrets. If you use GitBook, replace revoked API tokens and re-authenticate the MCP server.
  2. Cap and scope every AI key. Set a spend limit and the narrowest permission each key needs, so one leak does not become an unlimited bill.
  3. Pin packages and images. Use exact versions and digests, and delay adopting new releases by a day or two where you can.
  4. Lock down publishing. Use short-lived tokens, protect release workflows, and review any new workflow file a token pushes.
  5. Watch what agents can reach. Give agents narrow cloud roles, no standing production keys and a log you actually read.
  6. Patch WordPress, Apple devices, SonicWall, Cisco BroadWorks, Check Point Gaia, UnrealIRCd, IBM Aspera Faspex, FreeSWITCH, GPUStack, Barracuda ESG, Outlook, Chrome, React Server Components, N-able N-central, the llm CLI, fastapi-code-generator, kaniko, node-opcua, Kubeflow Pipelines, Claude Code, OpenSSH, Rails Active Storage, Aspire Dashboard and AWS mcp-gateway-registry, and revoke agent tokens. Update WordPress to 7.1.4 or later. Install current Apple, Outlook and Chrome updates. Follow Cisco advisory cisco-sa-bworks-xxe-uwUd7CEt for BroadWorks. Install the current Check Point Gaia and UnrealIRCd fixes. Update the llm CLI to 0.37.2, and apply N-able's Hotfix 2 for N-central. For SonicWall SonicOS SSLVPN, follow advisory SNWLID-2026-0014. Restrict the Barracuda ESG Admin UI. Move Aspera Faspex to 5.0.19 PL1, FreeSWITCH to 1.11.3, GPUStack to 2.2.2, fastapi-code-generator to 0.8.1, kaniko to 1.25.20 or later, node-opcua to 2.180.0, Kubeflow Pipelines to 2.18.0 and Claude Code to 2.1.251 or later. Move Rails to 8.0.5.2 or 8.1.4.1, Aspire Dashboard to 13.6.5 and AWS mcp-gateway-registry to 1.35.0. On musl systems such as Alpine, move OpenSSH Portable to 10.6p2 now, because it is exploited. If an infostealer may have run on a machine, revoke its Claude Code and Codex tokens, because a password reset does not do it.

How to follow new incidents

AgentBeam's November 2026 attacks tracker is updated every 4 hours. It lets you filter by type, gives every incident its own link, and has an Embed button under each entry so you can place a single incident on your own site or status page. A new tracker starts on the first of each month.

FAQ

How many cyberattacks happened in November 2026?

AgentBeam's November 2026 attacks tracker held 48 dated incidents, 6 unconfirmed claims and 45 earlier campaigns on November 20, 2026, because the month had just begun. Seven dated incidents are breaches, eighteen are critical vulnerabilities, twelve are exploited bugs, four are agent incidents, three are AI-run attacks and four are supply chain attacks. This post is updated daily as incidents are confirmed.

What were the biggest attacks of November 2026?

So far, four of the forty-eight dated incidents are from November 2. Kaspersky reports an LLM-run C2 server taking over education domains in 3 to 4 days. Northflank told all customers to rotate secrets, GitBook revoked all personal API tokens, and WorkOS confirmed access to an internal sales app. Arctic Wolf also confirmed AntiFox agents moving from IIS webshells to AWS secret reads in about 30 minutes. CISA added twelve bugs to its exploited list, including WordPress Core, Apple, SonicWall, IBM Aspera Faspex, FreeSWITCH, GPUStack, Barracuda ESG, Outlook, Chrome, Cisco BroadWorks, Check Point Gaia and UnrealIRCd flaws. For the largest incidents of the month before, see our October 2026 list, which counts 106 dated incidents.

What do most attacks have in common?

The sample is small, but many November incidents involve AI: three AI-run attacks, an attack on GitBook said to use AI agents, unauthorized use of customers' AI keys, infostealers that take Claude Code and Codex tokens, and a claim about a Railway engineer's AI agents. In October 2026, most attacks began with stolen credentials or trusted update channels, and many went after AI provider keys.

What should I do first after reading this list?

Rotate and scope any secrets you have not rotated since the October incidents, pin package versions, and cap every AI provider key by budget and permission so one leak cannot run up an unlimited bill.

How can I follow new incidents as they land?

AgentBeam's November 2026 attacks tracker is updated every 4 hours. Each incident has its own link, and each incident can be embedded on your own site with the Embed button under the entry. A new tracker starts each month.

Related reading

  • Fake OpenAI, SpaceX and Anthropic event invites phish Microsoft and Google logins
  • October 2026 cyberattacks: the full list of 106 incidents
  • The alleged Anthropic Mythos checkpoint on Hugging Face looks fake
  • The Internet Is Under Attack: protect yourself and the SaaS question
  • What is a data breach in the age of AI agents
  • Chainalysis breach: password hashes and live tokens for all customers

Figures and statuses are accurate as of November 20, 2026, and the underlying tracker keeps changing. Check the tracker for the latest.

Spotted something out of date? Let us know.
Daily briefFrom Yash · 350K+ students taught

Catch up on AI

One AI story explained in plain English, plus one thing to try. Every day, free. Unsubscribe anytime.

  • ✓ Free
  • ✓ One email a day, no spam
  • ✓ Unsubscribe anytime
Yash Thakker

Written by

Yash Thakker

Yash is an AI expert with over 300K learners. Join his workshops →

View Yash Thakker in People in AI →

Related posts

Oct 23, 2026

October 2026 Cyberattacks: The Full List of 106 Incidents, Counted by Type

October 2026 had 106 dated cyberattacks and security incidents from October 1 to 31. Nineteen were breaches, twenty were AI-run attacks, fourteen were agent incidents and thirteen were supply chain compromises. This is the complete list in date order, with the patterns behind it and the steps that matter now.

Nov 16, 2026

Google GTIG exploit_agent: First Fully Autonomous Patch-to-Exploit Pipeline

On November 16, 2026 Google Threat Intelligence Group reported the first observed fully autonomous vulnerability pipeline. A staging server hosted exploit_agent, wired to frontier and local LLMs, comparing web app code before and after patches, writing exploits, and executing them with Russian-language operator notes.

Nov 10, 2026

Hugging Face Warns Cyberattacks Could Make Open AI Hubs Less Open

On November 10, 2026, TechCrunch quoted Hugging Face chief policy officer Irene Solaiman warning that cyberattacks could push open AI platforms toward identity gates and upload restrictions. The piece ties OpenCode API key theft, this summer's agent breach, and Fireworks' weights incident into one supply-chain story.