Every deck is built from one kill chain page on this site: the same stages, the same MITRE ATT&CK techniques, the same figures and the same sources. A script checks each deck's figures, techniques and sources against its page, and each deck is reviewed against the page before it is published. The slides are free to share and adapt for non-commercial use under CC BY-NC 4.0.
Decks are published in batches, newest kill chains first. The list below grows as each one passes its checks, and the kill chain page links its deck at the same time.
On this page
The decks
Each card links the PDF, an editable copy and the kill chain the deck comes from. You can page through each deck in its card; the preview and the PDF show the slides without the speaker notes.
32 decks so far, newest incident first.
ARTEX / Exposed Secrets
Microsoft links GitHub scanning for secrets to ARTEX; in a follow-on attack, app credentials of unresolved origin reached Key Vault, SharePoint and MSPs' clients, per Microsoft and Todyl.
Binance / TraderTraitor
Per Binance's initial findings, a campaign that began months earlier reached an internal wallet signing service; about $6.3 billion left its hot wallets, and the FBI attributes the theft to North Korea's TraderTraitor.
Polymarket / DebugBear RUM Drainer
On October 29, 2026, a vendor's monitoring script on polymarket.com was swapped for a drainer that, per Polymarket, used traders' active sessions to sell positions and move funds.
WorkOS Escaped Research Agent
An agent reached a WorkOS internal sales tool through a public API and a predictable development credential; WorkOS now attributes it to a security researcher's ARTEX agent.
LiteLLM 1.107.0 Release Hijack
Per LiteLLM's attack-chain record, an admin key committed in May and never revoked led to an owner's GitHub token, a lookalike repository in the release workflow, and signed malware.
GMO Onamae.com Desktop Cloud
In GMO's account, an attacker got from a customer's remote desktop VPS to GMO's management server and pushed malware through it to customer desktops; saved broker logins were used for unauthorized trades, and extortion followed.
OVHcloud CLI Release Swap
Per OVHcloud, a developer's compromised GitHub account was used on October 19 to replace the binaries on five ovhcloud-cli releases with a stealer build; a customer's pinned hash in CI caught it.
VAPOR PANDA / ARTEX
AI agents, steered by a human at decision points, targeted 200+ organizations in 11 days, per CrowdStrike.
Storm-0940 / ConnectWise
Microsoft assesses that Storm-0940 was behind the ConnectWise Cloud Backup compromise and used ARTEX agents to attempt sign-ins to 2,250+ tenants; it says most of the targeting failed.
Black October / AI Developer Platforms
At least nine AI developer platforms reported security incidents in October 2026, and the recurring target was stored environment variables, especially keys that grant access to AI models.
M365 Token Spray and Replay
Huntress says an operator replayed stolen Microsoft 365 sessions, minted after MFA, from one rented /24, then registered rogue Entra devices, swapped MFA methods and hid mail.
CHAINDROP / keyv npm Worm
Attackers compromised the keyv maintainer; by August 6, per Elastic, 400+ npm packages were compromised, and the worm leaves a watcher that runs a remote-supplied handler when a stolen token is revoked.
Hugging Face / OpenAI Agent
OpenAI said in July 2026 that models it was evaluating escaped their sandbox and attacked Hugging Face's production infrastructure to get benchmark answers.
Storm-3168 / JADEPUFFER Azure Destruction
Per Microsoft, two compromised Azure service principals moved from hours of reads to 100+ storage deletion attempts in about seven minutes; how they were compromised is unconfirmed.
Suspected AI-Assisted AWS Compromise
Per Sygnia, one financially motivated actor went from an app weakness to broad control of a large AWS estate in roughly 72 hours; AI assistance is inferred, not proven.
Storm-2949 / Entra ID SSPR
Microsoft assesses with high confidence that Storm-2949 got in through social engineering consistent with known SSPR abuse; it then rode existing Azure roles to data theft.
Megalodon / GitHub Actions
On May 18, 2026, an actor pushed 5,718 commits that added or replaced secret-stealing workflows in 5,561 GitHub repositories; per Hudson Rock, 331 of 978 usernames tied to them matched infostealer-infected machines.
Mini Shai-Hulud / TanStack npm
On May 11, 2026 an attacker published 84 malicious @tanstack npm versions with valid SLSA provenance, after poisoning a GitHub Actions cache and reading OIDC tokens from runner memory.
Vimeo / Anodot
ShinyHunters obtained the tokens analytics vendor Anodot held for Vimeo and, within the integration's own permissions, queried Vimeo's Snowflake and BigQuery; Have I Been Pwned later counted about 119,000 unique email addresses in the leaked data.
Vercel / Context.ai
Per Vercel, a compromise of AI vendor Context.ai led through an employee's Google Workspace account into Vercel, exposing non-sensitive environment variables in a subset of customer projects; researchers trace it to an infostealer.
LiteLLM / TeamPCP
The maintainers traced it to a poisoned Trivy scanner in LiteLLM's own CI, which leaked the PyPI token; two malicious releases followed that swept cloud and AI provider credentials.
GTG-1002 / Claude Code
Per Anthropic, a Chinese state-sponsored group it calls GTG-1002 manipulated Claude Code into running the tactical work of an espionage campaign against about thirty organizations; the attribution rests on Anthropic's account.
Oracle EBS / Cl0p
Cl0p exploited CVE-2025-61882, an unauthenticated Oracle EBS flaw, from around August 2025, roughly two months before Oracle's October 4 patch; many victims first heard of it in an extortion email.
Shai-Hulud npm Worm
Beginning September 14, 2025, Shai-Hulud spread itself across more than 500 npm packages with no attacker involvement after the first push; how it began is disputed.
npm debug / chalk
An attacker phished one npm maintainer and relayed a live TOTP code; eighteen packages with over 2.6 billion weekly downloads carried a browser crypto drainer, pulled within hours.
Nx / s1ngularity
In August 2025, shell commands in a pull request title led to the theft of Nx's npm token, and the trojanized packages drove AI command-line tools installed on developers' machines to hunt for secrets.
Salesloft Drift / UNC6395
UNC6395 used stolen Drift OAuth tokens to bulk-export data from hundreds of Salesforce tenants, then combed it for secrets; Google estimated 700+ organizations were potentially affected.
Entra ID Actor Token
In July 2025 a researcher found that, through legacy Azure AD Graph, an Actor token from his own tenant allowed impersonating any user in any Entra ID tenant in Microsoft's public cloud; Microsoft found no evidence of exploitation before the fix.
UNC6040 / Salesforce Vishing
Google Threat Intelligence first reported UNC6040 in June 2025: callers posing as IT support talked employees into authorizing an attacker-controlled Data Loader, with no Salesforce vulnerability exploited.
tj-actions/changed-files
In March 2025, a stolen bot token let attackers repoint many tj-actions/changed-files version tags to one commit that printed CI secrets into public build logs.
Promptware
Researchers mapped a seven-stage kill chain onto attacks against cloud-connected AI assistants; most stages draw on one attack, from a Google Calendar invite to covert Zoom streaming.
Looking for a kill chain that is not listed yet? Its deck is still being written, and the full kill chain is there to read in the meantime. The how-to guides have decks of their own, linked from the top of each guide.
What every deck contains
The decks share one shape, so once you have presented one you can find your way around the others. A section is left out when its kill chain has nothing to put in it.
- A cover giving the date of the kill chain the deck reflects, then a slide on how to use the deck.
- At a glance: the headline figures, each with who reported it.
- What happened, with a timeline when the page gives dates, and a slide on who reported what when the sources disagree.
- The whole kill chain on one slide: every stage with its phase and its ATT&CK technique.
- One slide per stage, with its techniques, its key facts and why the stage matters.
- Background slides for concepts a stage relies on, such as XML external entities or IMDSv2. Each is labelled Background and names its own source, so you can tell general explanation from the facts of the incident.
- Defenses, starting with a matrix of which defense breaks which stage. A stage that no defense on the page addresses keeps an empty column, on purpose.
- Hunting markers, where the page gives them, with the page's caveats about how quickly they change.
- Check your own environment: the page's defenses turned into questions about your own estate.
- Discussion questions, each followed by an answer slide.
- Sources, related kill chains, and a closing slide that links the full page.
Every slide between the how-to slide and the closing slide has speaker notes written as a script for the presenter. The notes add narration and detail, and they never state a fact the slides contradict.
Presenting a deck
The notes are your script, so start there. Make your own copy (see below), open it in Google Slides and read the notes under each slide once through before you present. When you present, open the menu beside the Slideshow button and choose Presenter view: the room sees the slides and you see the notes.
Say the attributions out loud. A figure on a slide comes with its source ("per CrowdStrike") and its hedge ("about", "moderate-confidence"), and the notes repeat both. Dropped in the telling, a vendor's estimate becomes a fact you asserted.
Fit it to your slot
- For a short slot, like the 10 to 15 minute lightning format at the Friday session, keep the cover, At a glance, the kill chain on one slide, the stage slides and one discussion question.
- For a full session, present the whole deck, and decide first how much time the discussion questions get. That is where the room's own experience comes out.
- For a room that already knows a concept, skip its Background slide. In your copy, right-click the slide in the filmstrip and choose Skip slide: it stays in the deck, and the slideshow passes over it.
Run the discussion
Each discussion question has its answer on the next slide. Stop on the question, let the room argue, then move on to the answer, which says what the kill chain page says and gives the discussion somewhere to land. The defense matrix makes a good extra prompt: where a stage has an empty column, the page offers no defense for it, so ask the room what they would add.
Use it for team training
Run the Check your own environment slide as a working session. Each line on it is a question about your own estate, such as whether your instances require IMDSv2, and the questions nobody can answer become a short to-do list. A deck makes a ready-made opening for a tabletop exercise too: walk the stages, and at each one ask what your logs would have shown and who would have been paged.
Present it at CSOH
Want to walk the Friday session through one? Pitch it as you would any talk; Present at CSOH explains how.
Make it your own
The Make your own copy link asks you to sign in to a Google account, then saves an editable copy of the deck to your Google Drive. Edits stay in your copy and never touch the original. To work in PowerPoint or Keynote, open your copy and choose File, then Download, then Microsoft PowerPoint (.pptx).
Some ways teams adapt a deck:
- After each stage slide, add one showing how that step would look in your accounts: the log that would record it, the alert that would fire and the team that would respond.
- Where the deck lists the page's hunting markers, add the queries your team runs.
- Pair two decks that share an actor or a technique; where a deck has a Related kill chains slide, it names them.
- Cut the deck to the time you have, translate it, or move it onto your own slide template.
What the license asks
The decks, and the plain-text outlines they are built from, are licensed CC BY-NC 4.0. In short, you may share and adapt them for non-commercial purposes if you:
- keep the credit to Cloud Security Office Hours, csoh.org, which the cover and closing slides carry;
- link the license; and
- say that you changed the deck, for example with "Adapted from csoh.org, with changes" on your cover.
Non-commercial means not primarily intended for commercial advantage or payment, so selling the slides, or charging for a course built on them, falls outside the license. Material a slide quotes from a third party keeps its own license. The license text is the authority here; this section is a summary of it, not legal advice.
Keep the hedges when you edit
Every figure in a deck says who reported it and how sure they are, because the kill chain page does. When you shorten a slide, keep "per CrowdStrike", "about" and "moderate-confidence" with the number they qualify. Your copy still credits csoh.org, so a hedge dropped in an edit reads as the site's claim as well as yours; saying which slides you changed keeps the two apart.
Check the date on the cover
Breach pages are rewritten when new disclosures land, and a deck reflects its page as of the date on its cover. Before you present, compare that date with the kill chain page. If the page has changed since, trust the page.
Found something a slide gets wrong? Email admin@csoh.org or open an issue on GitHub, naming the deck and the slide number in its footer.
How the decks are made
Each deck starts as a hand-written outline, a plain-text file kept with the site's code on GitHub. Before it can be published, a script checks the outline against its kill chain page: every number written in digits, every ATT&CK technique and every source on a slide or in its notes must appear on the page, and every stage it names must exist there. Background slides are the one exception. They explain a general concept, say nothing about the incident beyond what the page says, and must name their own source instead.
The outline is then rendered into slides, reviewed slide by slide against the page, and shared from Google Slides. The same check runs whenever a kill chain page changes, so an edit that removes a figure a deck quotes fails the site's checks until the outline is fixed.
Where next
Read the breach kill chains the decks come from, the recurring root causes across them, or the year-by-year reviews. To run a deck as a session of your own, start with Present at CSOH.
Quick answers
Are the decks free to use?
Yes. Downloading the PDF, making your own copy and presenting a deck cost nothing. The decks are licensed CC BY-NC 4.0, so you can share and adapt them for non-commercial purposes as long as you credit csoh.org, link the license and say what you changed.
Do I need a Google account?
Only to make an editable copy. The PDF download and the preview on this page work without signing in.
Does the PDF include the speaker notes?
No. The PDF has the slides only, which is why every slide is written to make sense without its notes. To read the notes, make your own copy and open it in Google Slides.
Why does only part of the kill chain list have decks?
Each deck is checked against its kill chain page and reviewed before it is published, so decks arrive in batches, newest kill chains first. This page lists every published deck, and each kill chain page links its own deck once it exists.