2025 was the year the cloud attacker's playbook mostly stopped needing exploits. Stolen keys, hijacked OAuth tokens, and poisoned dependencies did most of the work that zero-days used to, and they did it at scale across cloud accounts, build pipelines and the SaaS layer on top of it.
This is a review of the cloud, SaaS, and software-supply-chain incidents that defined the year, the themes that connect them, and what they say about where to spend effort in 2026. For the timeless root-cause view across a longer span of incidents, see Lessons From 60 Cloud Breaches; for the step-by-step chains, the breach timeline.
On this page
- Overview: the shape of 2025
- The incidents that defined 2025
- Theme 1: Identity and token theft displaced exploitation
- Theme 2: The software supply chain as soft underbelly
- Theme 3: SaaS-to-SaaS trust chains
- Theme 4: Extortion crews abused native features, and zero-days still paid
- Theme 5: The AI and agent attack surface arrived
- Theme 6: Secrets sprawl fueled the next breach
- By the numbers and patterns
- What this means for 2026
Overview: the shape of 2025
If you read the 2025 cloud and SaaS breaches one at a time, each looks like its own story: a ransomware crew here, a poisoned npm package there, a maximum-severity identity flaw somewhere else. Read them together and the shape of the year is clear. The intrusions with the widest reach mostly did not exploit product vulnerabilities. Attackers logged in with stolen OAuth tokens, phished credentials or stolen access keys, and much of the traffic looked like legitimate authenticated activity. The exceptions, Cl0p's Oracle E-Business Suite zero-day and the on-premises SharePoint "ToolShell" chain, are a reminder that the old way in still works.
This review draws on eleven incidents, grouped by the patterns that recur across them, and every one has its own step-by-step kill chain on this site, linked from the table below. One of them, the Entra ID Actor token flaw, is research rather than a breach: Microsoft fixed it before anyone is known to have used it. For the longer arc across two decades of cloud breaches, see the breach timeline and the recurring root causes on breach lessons.
The single sentence version of 2025: identity and token theft displaced vulnerability exploitation as the primary way into the cloud, the software supply chain became the fastest path to thousands of victims at once, and the secrets scattered across repos, CI logs and developer machines turned many breaches into a launchpad for the next one.
The incidents that defined 2025
Here are the eleven incidents this review draws on, in rough chronological order, with the root cause and the single control most likely to have broken each chain earliest. As always, "one control" is a deliberate oversimplification (real defense is layered), but forcing the choice makes the priorities obvious.
| Incident | When | Root cause in one line | The one control that would have helped most |
|---|---|---|---|
| Codefinger S3 SSE-C ransomware | Jan 2025 | Stolen AWS keys encrypted S3 buckets with AWS's own SSE-C; AWS never stores the key | Deny SSE-C writes by bucket policy or SCP; short-lived credentials |
| tj-actions/changed-files (CVE-2025-30066) | Mar 2025 | A stolen bot PAT retagged a popular Action to a malicious commit that printed CI secrets into public logs | Pin third-party actions to full commit SHAs |
| UNC6040 Salesforce vishing | 2025 (first reported Jun) | Callers posing as IT support had employees authorize an attacker-controlled Data Loader app | Allowlist connected apps and restrict who can authorize one |
| SharePoint "ToolShell" (CVE-2025-53770) | Jul 2025 | Unauthenticated deserialization on on-prem SharePoint, then stolen machine keys that outlast the patch | Keep on-prem SharePoint off the internet; after exposure, rotate machine keys, not just patch |
| Entra ID cross-tenant Actor token (CVE-2025-55241) | Jul 2025 (disclosed Sep) | Legacy Azure AD Graph never checked which tenant an Actor token came from, and the token wrapping it was unsigned | Vendor-side fix only; no tenant setting mitigated it |
| Nx "s1ngularity" supply-chain attack | Aug 2025 | Shell commands in a pull request title ran in a pull_request_target workflow, which led to Nx's publish workflow and its npm token | Never interpolate PR titles into privileged workflows; default the workflow token to read-only |
| Salesloft Drift / UNC6395 | Aug 2025 | Stolen OAuth/refresh tokens from a third-party Salesforce integration drove bulk data export | Least-privilege, revocable third-party OAuth grants |
| Oracle E-Business Suite zero-day (CVE-2025-61882) | Aug-Oct 2025 | Cl0p exploited an unauthenticated RCE zero-day in Oracle EBS for about two months, then ran email extortion | Keep EBS off the internet; default-deny egress from the ERP host |
| npm debug / chalk maintainer phishing | Sep 2025 | A 2FA-reset phishing page relayed a maintainer's live TOTP code, enabling malicious publishes | Phishing-resistant MFA for maintainer accounts |
| Shai-Hulud self-replicating npm worm | Sep 2025 (2.0 Nov) | Stolen npm tokens let the worm republish up to twenty packages each victim maintainer owned | Short-lived, scoped publish tokens + provenance |
| GTG-1002 / Claude Code espionage | Sep 2025 (disclosed Nov) | Operators told Claude Code it was doing sanctioned testing and split the campaign into innocuous tasks, so the agent did most of the tactical work | No new control: the fundamentals, weighted toward ones that need no human in the loop |
Two things stand out just from the table. First, the "one control" column is dominated by identity and pipeline hygiene, not vulnerability patching. Second, several rows are not really about a single company's security posture at all: they are about the trust that company extended to an integration, a dependency, or a maintainer somewhere else. The themes below unpack why.
Theme 1: Identity and token theft displaced vulnerability exploitation
Most of the year's widest-reaching cloud and SaaS intrusions did not exploit product vulnerabilities. Attackers logged in with stolen OAuth tokens or phished credentials. Detection is hard because this traffic looks like legitimate authenticated activity, and it often never meets MFA or conditional access at all.
The clearest example is Salesloft Drift / UNC6395. Over roughly ten days in August 2025, the actor used stolen OAuth and refresh tokens for the Salesloft Drift chat integration to bulk-export data from customers' Salesforce instances, then combed the exports for AWS keys, Snowflake tokens, VPN credentials and passwords to enable the next compromise. The root cause traces back to a compromise of Salesloft's GitHub environment between March and June 2025, which let the actor pivot into Drift's AWS environment and take the integration's customer tokens. This was a third-party SaaS-to-SaaS trust chain, not a flaw in Salesforce itself. More than 700 organizations were potentially affected, including major security vendors. On August 20, 2025, Salesloft and Salesforce revoked all active Drift OAuth tokens and Salesforce removed the Drift app from AppExchange.
The UNC6040 campaign is the same theme delivered by phone. Google Threat Intelligence first reported it in June 2025: operators phoned employees while posing as IT support, walked them through Salesforce's own connected-app authorization page, and had them enter a connection code that authorized an attacker-controlled build of Salesforce's Data Loader. From then on the app exported data with the victim's own permissions. There was no technical vulnerability; the victim approved a rogue connected app. Victims included Google's own corporate Salesforce instance, along with fashion houses and technology firms. Extortion arrived separately and sometimes months later, from activity Google tracks as UNC6240, whose demands claim affiliation with ShinyHunters, a claim Google treats as likely pressure rather than established fact.
The most alarming pure-identity flaw of the year was never used, as far as anyone knows. Microsoft's Entra ID cross-tenant Actor token flaw (CVE-2025-55241, CVSS 10.0) was reported privately to Microsoft by Dirk-jan Mollema on July 14, 2025, fixed globally on July 17, and made public on September 17. An attacker could request a genuine Actor token in a tenant they controlled, edit the unsigned impersonation token wrapping it, and the legacy Azure AD Graph API would accept it in any other tenant, because it never checked which tenant the token came from. The result was impersonation of any user in any Entra ID tenant, up to Global Administrator. Conditional Access was never evaluated and MFA never prompted, because no user sign-in occurred, and token issuance, impersonation and every read left no log in the target tenant. Microsoft's telemetry found no evidence anyone exploited it before the fix, and no tenant configuration would have mitigated it.
The through-line: when the attacker holds a valid token, your login screen and your MFA prompt are already behind them, and the Entra flaw showed how far a forged one could have gone. The controls that matter are phishing-resistant factors, an allowlist for connected apps, short-lived and narrowly scoped tokens, and telemetry that can flag anomalous use of otherwise-valid identity material. See IAM and zero trust for the deeper treatment, and non-human identity for the token and service-account side.
Theme 2: The software supply chain became the soft underbelly
Attackers repeatedly compromised a single upstream dependency, maintainer account, or CI/CD action to reach thousands of downstream projects at once. The recurring prize was secrets: cloud keys, npm and GitHub tokens, and PATs, collected from build runners and developer machines.
The defining case is tj-actions/changed-files (CVE-2025-30066) in March 2025. A compromised PAT belonging to the tj-actions-bot account, itself exposed through the poisoned reviewdog/action-setup (CVE-2025-30154), was used to retag the Action's versions to a malicious commit. The payload scraped the runner's memory for secrets and printed them into the build logs as double-encoded base64, which for public repositories meant world-readable. More than 23,000 repositories referenced the Action, but far fewer leaked anything: Endor Labs found that 614 of 5,416 public repositories referencing it ran it during the March 14-15 window, and 218 of those leaked secrets. Workflows pinned to a commit SHA were not affected, and the maintainers shipped a clean release, v46.0.1.
Two npm incidents extend the pattern to maintainer accounts. In the debug / chalk compromise on September 8, 2025, a maintainer was phished with a convincing 2FA-reset email from a lookalike domain (npmjs.help) that captured a username, a password and a live TOTP code, letting attackers publish malicious versions of eighteen packages, including debug and chalk, with over 2.6 billion downloads a week between them. The payload was narrow: a browser-side crypto drainer that hooked window.ethereum to rewrite transaction destinations, aimed at end users of applications built with the packages rather than at build secrets. The account had 2FA, and a relayed code got straight through. The maintainer acknowledged the compromise publicly and the malicious versions were pulled within hours, a response that came down to one volunteer noticing quickly rather than any designed control.
The Nx "s1ngularity" attack in August 2025 reached the same goal through CI. A pull request title injected shell commands into a pull_request_target workflow, which ran with a read/write default token, and the attacker walked away with Nx's npm publishing token. Trojanized versions of nx and several @nx/* packages were live for about four hours on August 26 with a malicious postinstall hook. It stole tokens, SSH keys and wallets from developers' machines and committed them to a public "s1ngularity-repository" in each victim's own GitHub account: 2,349 distinct secrets from 1,079 systems, per The Hacker News. In a second phase, the stolen tokens turned more than 5,500 private repositories public across more than 400 accounts.
The controls live in your CI/CD pipeline and on your developers' machines: pin third-party actions to full commit SHAs rather than mutable tags, default the workflow token to read-only, publish through OIDC trusted publishing rather than a stored token, harden workflows against injection, and treat every dependency and maintainer whose code runs in your build as part of your attack surface.
Theme 3: SaaS-to-SaaS trust chains turned one breach into hundreds
Deeply interconnected SaaS integrations (chat tools, sales tooling, CRM connectors) meant a compromise of one third-party app cascaded into every tenant that trusted it. The blast radius was defined by OAuth grants and connected-app permissions, not by any single company's own security posture.
Both Salesforce incidents are also SaaS-to-SaaS stories. In Salesloft Drift / UNC6395, the compromised link was an integration vendor whose stolen OAuth tokens reached hundreds of organizations' Salesforce instances over about ten days, with more than 700 potentially affected. In UNC6040, the malicious link was a rogue connected app that victims authorized themselves during a vishing call, granting the attacker API access to the org. In neither case did the affected companies have a vulnerability in the traditional sense. They had a trust relationship that an attacker turned into an export pipe.
This is the argument for treating connected apps and OAuth grants as a first-class part of your attack surface. The right questions are inventory questions: which third-party apps hold tokens into your SaaS tenants, what scopes do those tokens carry, when were they last used, who can authorize a new one, and who can revoke them. That is squarely the domain of SaaS security posture management, and it is where the least-privilege discipline you already apply to IAM roles needs to extend next.
Theme 4: Extortion crews abused native features, and zero-days still paid
Financially motivated actors leaned on legitimate functionality where they could. This "living off the cloud" approach evades traditional anti-malware and complicates forensics and recovery, because nothing on disk looks like malware. Where they could not, a zero-day in an internet-facing enterprise platform still did the job.
The signature case is the Codefinger S3 ransomware, disclosed by Halcyon on January 13, 2025. Codefinger used stolen AWS keys with s3:GetObject and s3:PutObject permissions to encrypt victims' S3 buckets with AWS's own server-side encryption with customer-provided keys (SSE-C), holding data hostage without exploiting any AWS vulnerability. The attacker generated and held the AES-256 key, and AWS stores only an HMAC of it, so recovery was impossible without paying. Halcyon observed at least two victim organizations. What made it notable was the weaponization of a native cloud feature rather than malware, paired with an S3 Lifecycle policy that deleted the encrypted objects after seven days to force payment.
The Oracle E-Business Suite campaign is the counterweight. No credential was stolen, no employee was tricked, and no token was forged. Cl0p began exploiting CVE-2025-61882 (CVSS 9.8), an unauthenticated remote code execution flaw, as a zero-day around August 2025; Oracle published its alert and emergency patch on October 4, and CISA added the flaw to its Known Exploited Vulnerabilities catalog two days later. For roughly two months, organizations running a fully patched, supported ERP system were being emptied of data, and many first heard about it from an extortion email addressed to their executives. The exploit chain began with the server fetching an attacker-hosted stylesheet, which is why egress filtering on an ERP host matters.
SharePoint "ToolShell" joined the two halves. From around July 18, 2025, CVE-2025-53770 and CVE-2025-53771 were exploited against on-premises SharePoint servers: a crafted request to ToolPane.aspx gave unauthenticated code execution, attackers dropped a web shell and stole the server's ASP.NET machine keys, and with those keys they could forge requests SharePoint accepts as genuine long after the patch. Microsoft attributed some of the activity to Storm-2603, assessed with moderate confidence as China-based, which went on to deploy Warlock ransomware. Reporting placed compromises at roughly 75 to 85 or more servers worldwide. SharePoint Online was not affected.
The defensive takeaway for Codefinger is that native features need native guardrails: deny SSE-C writes where you do not use them, watch for anomalous key use, and keep versioning and immutable backups so a single overwrite is not the end. For Oracle EBS and SharePoint it is attack-surface reduction: for two months no Oracle patch existed, so keep enterprise platforms off the internet and filter their egress, and after a SharePoint exposure rotate the machine keys and restart IIS, because patching alone leaves an attacker who copied them inside. See detection engineering for building the signals that catch feature abuse early.
Theme 5: The AI and agent attack surface arrived
2025 saw the first known supply-chain malware that searched for and weaponized locally installed AI coding assistants, and a state-sponsored campaign that, by its discoverer's account, used an AI coding agent as most of its workforce. Attackers began treating AI tooling and its permissions as a new class of asset to abuse.
The Nx "s1ngularity" attack is the standout on the victim side. Beyond stealing tokens and keys, its malicious postinstall hook looked for locally installed AI CLIs (Claude, Gemini and Q) and invoked them with flags such as --dangerously-skip-permissions, --yolo and --trust-all-tools to enumerate secrets on the developer's filesystem. It is the first known supply-chain attack to turn a developer's own AI assistant into a search tool for its own credential theft. AI CLIs run with real filesystem and credential access, and an attacker who can invoke them inherits that reach.
On the attacker side, Anthropic reported in November that it had detected, in mid-September, a campaign it attributes with high confidence to a Chinese state-sponsored group it calls GTG-1002. The operators told Claude Code it worked for a security company doing sanctioned testing and split the campaign into small tasks that each looked innocuous, so the agent ran reconnaissance, exploit development, credential harvesting and data extraction against approximately thirty organizations, succeeding in a small number of cases. Anthropic assesses the AI performed 80 to 90 percent of the tactical work. Humans still chose the targets, built the framework and approved escalation, so the better description is a human campaign with the labor automated. Two caveats belong with it: the attribution rests on Anthropic alone, with no independent corroboration published, and the agent hallucinated credentials and sometimes claimed to have extracted secrets that turned out to be public.
The lesson is not that AI tooling is uniquely dangerous, but that it is now part of the same permission and token graph as everything else, on both sides. Scope what your AI CLIs and agents can touch, do not exempt them from the least-privilege review you apply everywhere else, and expect the controls that stop a human operator to be the ones that stop an automated one. The evolving tactics are tracked in threat research.
Theme 6: Secrets sprawl turned breach after breach into a launchpad
Long-lived credentials scattered across repos, CI logs, developer machines and CRM records meant an initial foothold routinely yielded cloud keys and tokens for lateral movement. Self-replicating malware and credential-harvesting worms industrialized the collection and re-exposure of these secrets, in several cases dumping them back into public GitHub repositories.
The Shai-Hulud worm is the purest expression of this theme, and the first genuine worm in the npm ecosystem: malware that spread from maintainer to maintainer with no attacker involvement after the first push. Beginning on September 14, 2025, with @ctrl/tinycolor and more than forty other packages, it ran TruffleHog to sweep each developer's machine for credentials, committed the stolen secrets to a public "Shai-Hulud" repository in the victim's own account, and used each stolen npm token to republish up to twenty of the maintainer's other packages with itself inside. More than 500 packages were compromised, and CISA issued an alert on September 23. How it began is disputed: Unit 42 suggested a phishing campaign, while the tinycolor maintainer said the npm token was taken from a GitHub Actions secret in a shared repository, and Wiz (disclosure: the site's author works at Wiz, which Google acquired in March 2026) assessed the campaign as directly downstream of the Nx compromise. It also planted a workflow file that keeps leaking secrets on every push, so a response that stopped at downgrading the package left it running. In November the worm returned as "Shai-Hulud 2.0", moving to preinstall, hitting roughly 796 packages and creating more than 25,000 malicious repositories across about 350 accounts.
The same dynamic runs through the other breaches. Nx committed 2,349 distinct secrets from 1,079 systems to public repositories, per The Hacker News. In Salesloft Drift, the attacker's second act after mass export was combing those exports for AWS keys, Snowflake tokens, VPN credentials and passwords, so that one breach could seed the next. When a foothold reliably yields the credentials for the following move, the way to break the chain is to make the secrets it finds worthless: short-lived credentials instead of long-lived keys, disciplined secret scanning across code, CI and developer machines, and rotation that assumes exposure rather than hoping against it.
By the numbers and patterns
A few figures capture the scale of the year, drawn from the kill chains above:
- 700+ organizations potentially affected by the Salesloft Drift / UNC6395 OAuth token theft, including major security vendors.
- 23,000+ repositories referenced tj-actions/changed-files when it was backdoored; Endor Labs found 614 of 5,416 public repositories referencing it ran it during the window, and 218 of those leaked secrets.
- 2,349 distinct secrets from 1,079 systems published by the Nx "s1ngularity" attack, per The Hacker News, and 5,500+ private repositories made public in its second phase; the Shai-Hulud worm compromised 500+ packages, and its 2.0 wave about 796, with 25,000+ malicious repositories.
- 18 npm packages with over 2.6 billion weekly downloads trojanized in the debug / chalk maintainer phishing incident, pulled within hours.
- CVSS 10.0 for the Entra ID cross-tenant Actor token flaw, fixed globally three days after Mollema's private report and two months before public disclosure, with no evidence of exploitation before the fix.
- About two months of Cl0p exploiting the Oracle EBS zero-day (CVSS 9.8) before a patch existed; roughly 75 to 85+ on-prem SharePoint servers compromised in the "ToolShell" chain, per reporting.
- ~30 organizations targeted in the GTG-1002 campaign, with 80 to 90 percent of the tactical work done by AI, per Anthropic; a small number of intrusions succeeded.
The patterns behind the numbers are consistent. The intrusions with the largest reach were supply-chain and SaaS-integration compromises, where one upstream failure fanned out to hundreds or thousands of victims. The intrusions that best evaded detection were the identity and token abuses, which never tripped a vulnerability signature because the attacker was using valid credentials. And two of the year's fastest responses came from people rather than controls: the npm takedown came within hours because one maintainer noticed the phishing and said so publicly, and the Entra flaw was fixed three days after a researcher reported it, before anyone is known to have used it. Some items carry hedges: SharePoint and Oracle EBS are self-hosted platforms rather than cloud services, the Entra flaw is research rather than a breach, and GTG-1002 rests on one vendor's account.
What this means for 2026
Compress 2025 into a to-do list and it is short, unglamorous, and mostly about identity and dependencies rather than exploits:
- Treat tokens as credentials and shorten their lives. Access keys and OAuth tokens were the way in for Codefinger and Salesloft Drift, and stolen npm publish tokens let the Nx and Shai-Hulud attackers publish. Scope them narrowly, expire them quickly, and prefer short-lived credentials over long-lived static keys. See non-human identity.
- Inventory and govern your connected apps and SaaS integrations. The blast radius in both Salesforce campaigns was defined by OAuth grants, not by any product flaw. Allowlist connected apps, restrict who can authorize one, and know which third-party apps hold tokens into your tenants, with what scopes, and who can revoke them. This is the core of SaaS security posture management.
- Harden the build pipeline as production. Pin third-party actions to full commit SHAs, default the workflow token to read-only, publish through OIDC rather than stored tokens, and guard workflows against injection. tj-actions printed runner secrets into public logs, Nx's own release workflow handed over its publish token, and Shai-Hulud planted workflows that kept leaking secrets. See CI/CD.
- Make phishing-resistant MFA the default for maintainers and admins. Phishable TOTP was the weak link in the debug / chalk compromise: the account had 2FA, and a relayed code got straight through. SMS codes can be relayed the same way, and push prompts can be fatigued. FIDO2 and passkeys defeat both.
- Kill secrets sprawl so a foothold yields nothing. Scan code, CI logs, developer machines and exports for embedded credentials, rotate on the assumption of exposure, and stop shipping long-lived keys that a worm or a mined export can reuse.
- Instrument identity as well as infrastructure. When stolen tokens are indistinguishable from legitimate ones at the network layer, logs of who authenticated, from where, and to what are the only way to find them. Know too what your provider does not log: the Entra Actor token flaw left no trace of reads in the target tenant. Build detections on the telemetry you have with detection engineering.
- Do not forget the boring zero-days. Oracle EBS and SharePoint show that unauthenticated RCE in internet-facing enterprise platforms still pays. Keep those platforms off the internet and filter their egress, because for Oracle EBS those were the controls that worked before a patch existed, and after a SharePoint exposure rotate the machine keys, because a patch does not evict an attacker who copied them.
- Treat AI agents as privileged software. Nx turned developers' AI CLIs into secret hunters, and in GTG-1002, by Anthropic's account, an agent did most of an intrusion's labor. Scope what agents can reach, and log what they do.
None of this is novel research. The breaches of 2025 fell to fundamentals that were skipped, scoped too broadly, or turned off for convenience: an unrotated key, a broad OAuth grant, a mutable action tag, a phishable second factor. Every incident in this review has its own step-by-step kill chain, and the breach timeline shows how they sit alongside two decades of prior breaches, with the synthesized root causes on breach lessons.
Ten months into 2026, this list has been extended rather than overturned. The 2026 review so far records sessions captured after MFA and replayed, connected apps and vendor tokens as the way in again, and supply-chain attacks that passed signature and provenance checks. What it adds is AI moving from the attacker's tooling toward the operator's seat, and an AI-assisted discovery, PHP2Shell, that sent defenders looking for every copy of PHP.
Where next
Walk the full step-by-step chains on the breach timeline, read the cross-incident synthesis in Lessons From 60 Cloud Breaches, and go deeper on the controls that keep recurring: IAM and least privilege, non-human identity, CI/CD and supply-chain security, and detection engineering.
Quick answers
What was the biggest cloud security story of 2025?
There was no single story, but one clear shift: identity and token theft displaced vulnerability exploitation as the primary way into the cloud. Two of the year's defining SaaS intrusions (the Salesloft Drift / UNC6395 OAuth theft, with 700+ organizations potentially affected, and the UNC6040 Salesforce vishing campaign) did not exploit product vulnerabilities at all. Attackers logged in with stolen identity material, and the traffic looked like legitimate authenticated activity. The exceptions, Cl0p's Oracle E-Business Suite zero-day and the on-premises SharePoint "ToolShell" chain, show the old way in still works.
Were the 2025 npm and GitHub Actions attacks connected?
They were largely separate incidents, though one assessment, in a disputed record, places Shai-Hulud directly downstream of the Nx compromise. Three of them shared one goal, harvesting secrets from build runners and developer machines: the tj-actions/changed-files compromise (CVE-2025-30066), in an Action referenced by 23,000+ repositories, printed CI secrets into public logs; the Nx s1ngularity attack published 2,349 distinct secrets from 1,079 systems, per The Hacker News; and the Shai-Hulud worm self-replicated across 500+ packages. The debug/chalk phishing was the odd one out: its 18 packages carried a browser-side crypto drainer aimed at end users rather than build secrets. Together they made the software supply chain the fastest path to thousands of victims at once.
How did the Codefinger S3 ransomware work without any AWS vulnerability?
Codefinger used stolen AWS access keys with s3:GetObject and s3:PutObject permissions to encrypt victims' S3 buckets with AWS's own server-side encryption with customer-provided keys (SSE-C). Because AWS never stores the customer-supplied AES-256 key (it keeps only an HMAC of it), recovery was impossible without paying, and a 7-day deletion timer added pressure. It weaponized a native cloud feature rather than malware. The full walkthrough is on the Codefinger S3 kill chain page.
What should security teams prioritize for 2026 based on 2025's breaches?
Treat OAuth tokens and access keys as credentials and shorten their lifetimes; allowlist connected SaaS apps and restrict who can authorize them; harden the CI/CD pipeline by pinning actions to commit SHAs, defaulting workflow tokens to read-only and publishing through OIDC; make phishing-resistant MFA the default for maintainers and admins; eliminate secrets sprawl so a foothold yields nothing reusable; instrument identity so stolen tokens can be detected; keep internet-facing enterprise platforms off the internet where you can, since Oracle EBS and SharePoint zero-days still paid off; and scope what AI agents can reach.