Articles
- Collins Aerospace 2025: one check-in vendor's ransomware put four European airports on paper
On 19 September 2025 ransomware hit the check-in software that Heathrow, Brussels, Berlin and Dublin all rented from Collins Aerospace. Not an avionics hack: the ground software and a shared vendor network. Dated timeline, reported cancellation figures (217 counted, Brussels 140 of 276 requested), the end-of-life systems seen in public scans, what is still unconfirmed, and a one-hour vendor check for small companies.
- CDK Global 2024: 15,000 car dealerships stopped for two weeks and a reported $25 million payment
One hosted dealer management system went offline on 19 June 2024 and took about 15,000 dealerships with it. The dated timeline, the 387 bitcoin TRM Labs traced, the $1.02 billion dealer-loss estimate with the arithmetic, the impersonation calls that followed, and what CDK still has not said: how the attackers got in.
- Oracle E-Business Suite 2025: Clop's zero-day extortion emails and what to check on an internet-facing ERP
Clop's Oracle E-Business Suite campaign in dates: exploitation from 9 August, extortion emails from 29 September, the emergency fix on 4 October (a 56-day window). The URLs, database tables, implants and IP indicators to search for, and seven checks for an internet-facing ERP.
- Co-op 2025: shutting the network down before the encryptors ran, and what it cost anyway
The Co-op stopped ransomware in April 2025 but lost 6.5 million members' data and £285 million of revenue. The reset that started it, what its executives told MPs about segmentation, and the cost table from its own unaudited results, with the controls that break each step.
- GitLab 2024: CVE-2024-0402 arbitrary file write, account takeover and CI token exposure
On 25 January 2024, GitLab issued an emergency critical security release for CVE-2024-0402 (CVSS 9.9): an arbitrary file write vulnerability within GitLab Workspaces devfile processing. By exploiting insecure archive decompression in a Go helper binary, an authenticated attacker could escape workspace boundaries, overwrite host files like authorized_keys, achieve remote code execution, and extract CI runner tokens across developer repositories. The mechanics of the flaw, escalation paths, and defense blueprints for developer infrastructure.
- Marks & Spencer 2025: a help-desk reset, DragonForce and 46 days without online orders
M&S lost its online shop for 46 days in 2025 after an impersonation involving a third party on 17 April. What is established and what is only reported, the dates, the £131.3 million cost and £100 million insurance in M&S's own results, and the help-desk and segmentation controls that break each step.
- Uber 2022: a contractor password, repeated MFA pushes and one PowerShell file with admin credentials
On 15 September 2022 an attacker bought a contractor's password harvested by an infostealer, bombarded their phone with MFA pushes until a fake WhatsApp message got it approved, and found a PowerShell script holding master PAM admin credentials on an open network share. Step by step, how simple push MFA fails, why internal shares leak root keys, and how to audit and fix each step.
- Hot Topic and Snowflake 2024: a vendor's infostealer log, 350 million claimed, 57 million counted
A forum seller claimed 350 million Hot Topic customers; Have I Been Pwned counted 56,904,909 accounts. What Hudson Rock found on a vendor employee's infected computer, what was only the attacker's claim, what the public record still lacks, and the Snowflake checks to run on your own vendor logins.
- PowerSchool 2024: one support credential, a claimed 6,505 districts and two extortion rounds
A compromised subcontractor support account opened PowerSchool customers' SIS instances through a remote-maintenance feature. The attacker exported Student and Teacher tables, demanded about $2.85 million, and months later schools received fresh demands. The confirmed timeline, the unconfirmed 6,505-district claim and the controls that break the chain.
- New York AG's 2022 credential stuffing report: 1.1 million accounts at 17 companies and the controls it asks for
In January 2022, the New York Attorney General released a landmark investigation into credential stuffing: 1.1 million compromised accounts across 17 major retailers, restaurant chains, and food delivery services. Most companies were completely unaware of the attacks. The cross-firm findings, how mobile APIs and residential proxies bypassed standard controls, and the four-pillar safeguard blueprint mandated by the AG.
- DraftKings 2022: 60,000 betting accounts, a $5 deposit and $600,000 stolen
On 18 November 2022 a credential-stuffing ring opened about 60,000 DraftKings accounts. Buyers added a payment method, deposited $5 to verify it, then withdrew the victim's balance: about $600,000 from 1,600 accounts. The federal case, the 67,995-person breach filing, the money flow and the controls that would break each step.
- Dropbox 2022: a fake CircleCI login and 130 copied repositories
A CircleCI-themed email led a Dropbox developer to a lookalike login that collected GitHub credentials and a copyable hardware-token code. The attacker copied 130 repositories before GitHub raised the alarm. Here is the dated chain, what the repositories held, and the control that breaks each step.
- num2words 2025: Scavenger moves to PyPI after the npm phishing wave
On 28 July 2025, ten days after hitting npm, the Scavenger malware group targeted PyPI maintainers with spoofed verification emails from pypj.org. Malicious versions of num2words (7.8M monthly downloads) bypassed wheel install-time limits by injecting ctypes loader code into __init__.py. The timeline, real download numbers, and how pip and uv teams stay safe.
- Dunkin' DD Perks 2015-2019: 19,715 accounts, 300,000 more and New York's $650,000 order
A vendor told Dunkin' in August 2015 that 19,715 DD Perks accounts had been logged into with reused passwords; a second vendor flagged about 300,000 more in 2018-19. The five-day traffic sample (5.4 million attempts, 0.365% success), what stored value cards were worth, what the consent order makes a company do after a warning (30-day resets, 3-business-day histories, refunds without receipts, 5 years of records) and a control for each step.
- Norton LifeLock 2022: 6,450 password-manager accounts at risk after credential stuffing
A reused Norton login opened 6,450 customer accounts in December 2022, and Gen Digital secured 925,000 that may have been targeted. What those figures do—and do not—measure, why a matching vault key multiplies the damage, and the controls that break each step.
- eslint-config-prettier 2025: a phishing email from npnjs.com and the Scavenger loader in a lint config
On 18 Jul 2025 a spoofed npm email pointed a maintainer at npnjs.com. Nine versions in five packages (97 million weekly downloads) then shipped a Windows stealer at install time. The registry timeline, how Jest 30 could pull it in, and our test of seven package managers: which settings still run the script.
- Hot Topic 2024: a reported vendor infection, a claimed Snowflake login and 57 million exposed email addresses
An attacker claimed 350 million Hot Topic, BoxLunch and Torrid rows. Have I Been Pwned validated 56,904,909 unique email addresses. Here is what the evidence supports, what remains attributed reporting, and the controls that break the path from an infected vendor computer to a data warehouse.
- Chick-fil-A 2023: 71,473 app accounts, loyalty points and stored cards in a 57-day stuffing run, and a repeat in 2026
From 18 Dec 2022 to 12 Feb 2023 automated logins with reused passwords opened 71,473 Chick-fil-A One accounts: points spent, $50 charges on stored payment, emails swapped, accounts resold for $2 to $200. The dated timeline, attempt arithmetic, why points go so easily, why the settlement figure was never published ($4.4M is a different case), the 13,322-person repeat in June 2026 and a control for each step.
- PayPal 2022: 34,942 accounts opened by credential stuffing in two days, and a $2 million New York penalty
On 6-8 December 2022 attackers logged in to 34,942 PayPal accounts with passwords leaked elsewhere, and read full Social Security numbers off new 1099-K tax forms that had gone live unmasked seven weeks earlier. The NYDFS consent order (January 2025, $2 million) shows why: the change was filed as a platform migration and skipped review, MFA was optional, and there was no CAPTCHA or rate limit. Timeline, attempt arithmetic and the control for each step.
- chalk and debug 2025: a 2FA-reset email from npmjs.help and two hours of crypto-clipper code in 18 packages
On 8 Sep 2025 a 2FA-reset email from npmjs.help relayed qix's password and TOTP code to npm. chalk, debug and 16 more packages (2.67 billion weekly downloads) shipped a browser crypto clipper for about two hours. The dated timeline, the 18 versions, what the payload hooked, why only about $500 was taken, and the control that breaks each step.
- PAN-OS management interface 2024: an authentication bypass chained to root
Operation Lunar Peek targeted public PAN-OS management interfaces in November 2024. CVE-2024-0012 gave an unauthenticated attacker administrator privileges; CVE-2024-9474 then made command injection a root-level problem. Why this was not GlobalProtect CVE-2024-3400, what Palo Alto observed, and a firewall-by-firewall remediation plan.
- GoAnywhere MFT 2025: a forged license response and Medusa ransomware
Microsoft saw Storm-1175 exploit GoAnywhere MFT a week before public disclosure of CVE-2025-10035. A forged license-response signature reached attacker-controlled deserialization in the License Servlet; RMM tools, JSP files, discovery, Rclone and Medusa followed. The timeline, an evidence-led hunt, and how to contain an internet-facing MFT server.
- CrushFTP 2025: an S3 authorization header that logged attackers in as crushadmin
Two actively exploited CrushFTP authentication bypasses landed four months apart in 2025. The March flaw let an AWS4-HMAC-SHA256 Authorization header select a local username before the password check; July's AS2 validation flaw again exposed internet-facing file-transfer servers. What the timeline says, why 11.3.0 is still exposed to both, and the incident checklist.
- Twitter 2020: phone calls to staff, an internal admin panel and 130 famous accounts
In July 2020 teenagers phoned Twitter staff as the IT help desk, relayed their VPN passwords and MFA approvals in real time, and used an internal tool that over 1,000 employees could open to take 130 accounts, including Obama, Musk and Apple. The steps from the NYDFS report and court papers, the bitcoin count, and the controls an internal admin panel needs.
- TanStack 2026: a fork pull request that poisoned the pnpm cache, and why GitHub made caches read-only
On 11 May 2026 a fork pull request ran in TanStack's pull_request_target job, wrote a 1.1 GB poisoned pnpm store into main's Actions cache, and closed itself. Eight hours later the release restored it, read the OIDC token from runner memory and published 84 malicious versions. The timeline, the files as they were, what our Workflow risk check flags and misses, GitHub's read-only cache change, and a release layout that holds.
- A 3-day release cooldown: which malicious npm and PyPI releases of 2024-2026 it would have skipped
18 malicious releases of popular npm and PyPI packages, December 2024 to October 2026, with publish and removal times: the longest known window was 12.4 hours, so a 1-day or 3-day cooldown skips 17 of 18 (the 18th is a worm's long tail). The settings for Renovate, Dependabot, npm, pnpm, Yarn, Bun, uv and pip, and the real cost: a 3-day gate also holds back the security fix unless you exclude it.
- PAN-OS GlobalProtect 2024: a session cookie that wrote a file, and a firewall that ran it as a command
CVE-2024-3400 joined two small flaws on Palo Alto firewalls: GlobalProtect created an empty file wherever the session cookie pointed, and a telemetry job put file names into a shell command. Each step from Volexity, Unit 42 and watchTowr, the UPSTYLE backdoor that read commands from the error log, hotfix builds per branch, exposure counts, and what one firewall's owner should check and rotate.
- Okta and Lapsus$ 2022: a support contractor's laptop and five days of remote access
In January 2022 an attacker had remote access for five days to a Sitel support engineer's laptop that was signed in to Okta. Okta first said up to 366 customers, then two; customers heard two months late, from Lapsus$ screenshots. The timeline, what one support login could do, and the contract and access questions for any company that outsources support.
- Nx s1ngularity 2025: a workflow injection that published malicious npm packages and used AI command-line tools to hunt secrets
In August 2025 a pull request title ran as a command inside Nx's GitHub Actions, the npm token leaked, and eight nx versions shipped a postinstall that asked Claude, Gemini and Q to hunt for wallets and keys. The timeline, the 2,349 leaked secrets and 6,700 repos made public, the control for each step, and our test of npm 12's install-script blocking.
- MOVEit 2023: one SQL injection, 2,773 organisations and 95 million people
In May 2023 the Cl0p group used a SQL injection zero-day in MOVEit Transfer to drop a web shell and pull files from hundreds of servers, then extorted the owners without encrypting anything. Most of the 2,773 organisations hit never ran MOVEit: their vendors did. Step by step, and the controls that break each step.
- Mailchimp 2023: staff tricked into giving credentials, an internal admin tool and 133 customer accounts
In January 2023 phished Mailchimp employees and contractors gave an intruder the internal support and account-admin tool, and 133 customer accounts. It was the third time in ten months: 319 accounts in March 2022, with Trezor phishing after, and 214 in August. The three incidents side by side, why support tools are the target, a view-count alert with the query, and controls for both sides.
- Ledger Connect Kit 2023: a former employee's npm account and five hours of wallet drainers on DeFi sites
On 14 December 2023 a phished former Ledger employee's npm access, never revoked, published three malicious versions of @ledgerhq/connect-kit. A loader that fetched '@1' from jsDelivr at runtime put the drainer into SushiSwap, Zapper, Revoke.cash and other dApps that never redeployed. The minute-by-minute timeline from the npm registry, the figures (about $600k), and the controls that break each step: offboarding, token rules, exact versions with SRI, trusted publishing.
- event-stream 2018: a popular npm package handed to a stranger, and a wallet stealer inside
In 2018 the author of event-stream, downloaded 2.2 million times a week, handed it to a stranger who asked by email. A new dependency, flatmap-stream, carried an encrypted payload that only woke up inside the Copay wallet's build and stole large Bitcoin wallets. The registry timeline, each step and its control, and a check we ran on our own lock file for first releases by new publishers.
- Coinbase 2025: support agents paid to look up customers, then the 'Coinbase security' phone calls
In 2025 overseas support agents were paid to look up Coinbase customers and photograph what the support tool showed. 69,461 people's records fed phone calls from fake 'Coinbase security' that talked customers into moving their own funds. A $20M demand refused, $311M paid out. The chain step by step, a field-by-ticket table, a lookup alert query, and why the customers' own 2FA didn't matter.
- Bybit 2025: one developer laptop, a swapped JavaScript file in a storage bucket and a $1.5 billion transfer
In February 2025 a Safe{Wallet} developer's laptop led to AWS session tokens, then to one swapped Next.js file in the S3 bucket behind app.safe.global. The file acted only for Bybit's cold wallet, and $1.46 billion left in one signed transaction. Step by step, with our hashes of the archived file before and after, why CSP allowed it and SRI would not have, and the controls for any front end served from a bucket.
- Authy 2024: an open API endpoint that confirmed 33 million phone numbers
In June 2024 ShinyHunters fed lists of phone numbers to an Authy API endpoint that needed no sign-in and kept the 33,420,546 that came back as accounts. No account was taken over, but the list is a ready target for SMS phishing and SIM swaps. How account-existence checks leak, the math on why per-IP limits fail, Facebook, Twitter and Trello side by side, and a lookup design with an alert query.
- ArcaneDoor: Cisco firewalls backdoored below the operating system, three times
The group behind ArcaneDoor went after Cisco ASA and Firepower firewalls in 2024, 2025 and 2026, each time lower or longer-lived: a Lua script on the flash disk, a bootkit in the boot ROM of old 5500-X boxes with a loader that rebooted on a core dump, then an implant that survived the patches until the power was pulled. The implants, CISA ED 25-03's steps and dates, the models that can't be cleaned, fixed builds, and the checks for one ASA.
- Card skimmers inside Google Tag Manager containers: how a store's own trusted tag became a checkout stealer (2021-2026)
Since 2021 Magecart groups have hidden card skimmers in Google Tag Manager containers: 316 stores in the first year, 165,000+ cards, 3.5 months on average before removal. Sucuri's 2025 Magento case kept the tag in a database block; Sansec's 2026 one loads the skimmer from Stripe and sends cards back to Stripe. Why CSP allowlists and SRI don't catch it, and the checks that do.
- Akira and SonicWall SSL VPN 2025: passwords carried over in a Gen 6 to Gen 7 migration, OTP accounts still falling, encryption in under four hours
SonicWall looked at fewer than 40 Akira intrusions and found no zero-day: CVE-2024-40766 from 2024, and SSL VPN passwords nobody reset, many carried from Gen 6 to Gen 7 in a settings import. Arctic Wolf's figures (scan in 5 minutes, encryption in under 4 hours, OTP accounts hit), the open MFA question, and the controls that break each step.
- Watch your firewall's version: a Firmware version watch email, read line by line
A product guide to the osec Firmware version watch: save a firewall or VPN version from a Firmware branch check, what makes it send an email (a new CISA KEV entry or a new NVD mapping), a real email for NetScaler 14.1-43.56 read line by line, what to do after one, and what the watch costs on your quota.
- SonicWall SMA 100 2025: a rootkit on fully patched appliances, put back with stolen admin credentials
In June 2025 an attacker signed in to a fully patched SonicWall SMA 100 as its admin with a password stolen months earlier, opened a shell, and installed OVERSTEP: a rootkit loaded into every process through ld.so.preload, rebuilt into the boot image on each restart, with its commands hidden in web log lines. Each step from GTIG's report, the CVEs and dates, the end of support brought forward 23 months, and what to rotate and check.
- MySonicWall 2025: every customer's cloud-stored firewall configuration backup accessed, and what an encrypted backup still gives away
SonicWall first said under 5% of firewalls; on 8 October 2025 it said every customer who used MySonicWall cloud backup had their configuration file taken, through an API call, by a state-sponsored actor. What the file holds in plain view, what breaking AES-256 or 3DES would cost (and why that's the wrong question), and the 29-item reset in an order that won't lock you out.
- Storm-2372 2025: device code phishing that signed victims in for the attacker
From August 2024 Storm-2372 sent fake Teams invites whose meeting ID was a Microsoft device code. Victims typed it at microsoft.com/devicelogin, passed MFA, and handed access and refresh tokens to the attacker, who then registered a device in Entra ID. How the flow works, the 15-minute window, and the Conditional Access block that stops it.
- Disney 2024: a fake AI art tool on an engineer's PC and 1.1 TB of internal Slack
A Disney engineer installed a free ComfyUI add-on from GitHub on his home PC. Trojanised openai and anthropic packages in its requirements.txt stole his browser passwords and a password manager without 2FA; in May 2024 the attacker used them to download about 1.1 TB, 44 million messages, from Disney's Slack, then leaked it as "NullBulge". The chain step by step, what leaked in numbers, why 2FA and MFA alone fall short, and the control for each step.
- Roku 2024: two credential stuffing waves, 591,000 accounts and why Roku made two-factor mandatory
Between December 2023 and April 2024 attackers logged in to about 591,000 Roku accounts with passwords leaked elsewhere: 15,363 in a first wave, about 576,000 in a second. Fewer than 400 were used to buy hardware and subscriptions with stored cards; accounts were resold from 50 cents. The tools, the attempt arithmetic, why per-IP limits missed it, how Roku's email-link 2FA works, and the control for each step.
- Is your FortiOS 7.2 still exposed? Reading a Firmware branch check result
A product guide to the osec Firmware branch check, on real results from 26 Oct 2026: FortiOS 7.2.11 and CVE-2025-68686 (no fix on 7.2, first fix 7.4.7), Ivanti R-builds, PAN-OS hotfix builds, what still applies / to check / not mapped mean, and the clean-up steps an update doesn't do.
- Behind a CDN is not behind a firewall: CloudFront, Vercel and what a scan can see
4 of the 13 sites with no firewall in our check of 19 small sites sat behind a CDN or host edge. A CDN caches and speeds up; a WAF decides which requests reach your app. How to tell which one you have, from the DNS, the headers and one harmless test request.
- We checked 19 small business and developer sites: 13 had no firewall in front
We ran our site protection checks over 19 small sites in the USA and India: inns, restaurants, freelancers and small studios. 13 showed no firewall, 5 sent none of six basic security headers, and only 4 had a CSP. The data, what each kind of host can do about it, and how to check your own site in a minute.
- Ticketmaster 2018: a chatbot vendor's script on the payment page
Ticketmaster UK loaded a support chat bot from Inbenta on every page, including checkout. From 10 February 2018 the vendor-hosted script carried a card skimmer, and it ran for 133 days, 78 of them after banks first warned. The ICO fined Ticketmaster £1.25 million. The timeline from the penalty notice, how it differs from the BA case, what PCI DSS 6.4.3 and 11.6.1 now ask, and a payment-page checklist with SRI and CSP examples.
- Retool 2023: a text message, a deepfaked voice and synced authenticator codes
In August 2023 one Retool employee answered a fake IT text, then a phone call in a deepfaked familiar voice, and handed over two one-time codes. The attacker added a device to their Okta account, and because the employee's codes were synced to a Google account, one Google session opened all of them. 27 crypto customers' accounts were taken over. Step by step, why synced seeds make MFA one factor, and what phishing-resistant sign-in changes.
- Ultralytics 2024: a poisoned PyPI release built by the project's own GitHub workflow
In December 2024 a stranger's pull request put a shell command in its branch name. Ultralytics' pull_request_target workflow pasted it into a script, the attacker used the runner's token to poison the pip cache, and the project's own release workflow built 8.3.41 with a crypto miner and published it through trusted publishing, attestation and all. Then an old API token shipped two more. The times, the cache rules, and the settings that break each step.
- FortiGate 2025: a symlink that kept attackers reading files after the patch
Attackers who broke into FortiGate firewalls through three SSL-VPN flaws (2022-2024) left a symbolic link in the folder that serves the VPN portal's language files. Upgrades fixed the bugs and kept the link, so the configuration stayed readable without a login: up to 28 months for a promptly patched box. 16,620 devices still had it in April 2025, and the clean-up was bypassed in 2026. Step by step, the dates, and what to replace.
- Is your GitHub repo one pull request away from a poisoned release? Reading a Workflow risk check result
Each finding the osec Workflow risk check shows, on the real files behind the Ultralytics, tj-actions and Nx incidents: injected PR text, pull_request_target checkouts (also through a step output), composite actions, release jobs that restore the cache, missing permissions, unpinned actions. The fix to paste for each, and a 10-minute repo settings pass.
- PyTorch nightly 2022: a package name on the public index that beat the private one
Over Christmas 2022 a stranger uploaded torchtriton 3.0.0 to PyPI, a name PyTorch only hosted on its own index as 2.0.0. pip merged both indexes, took the higher number, and Linux nightly installs ran a binary that sent SSH keys and home files out over DNS. The timeline, our own pip and uv test of the two-index rule, and how a small team scopes its package sources.
- Pin every GitHub Action to a commit in two minutes with the osec Action pin lookup
A uses: line with @v4 follows a tag that can be moved to other code, which is how tj-actions/changed-files leaked secrets in 2025. Paste your uses: lines into osec's free Action pin lookup, read the flags, paste back commit SHAs with version comments, and let Dependabot keep them current.
- Stop piping curl into a shell: pin an install script in two minutes
curl | sh runs whatever the server sends that day. Paste the install URL into osec's free Pinned download check, read the notes, and copy a download-verify-run snippet for shell, GitHub Actions, GitLab CI or a Dockerfile. Plus what to do when the hash changes.
- Dropbox Sign 2024: one automated service account opened the customer database
In April 2024 an intruder used a back-end service account of Dropbox Sign (formerly HelloSign) to reach data on every user: email addresses, plus hashed passwords, API keys, OAuth tokens and MFA details for some. No person's login was involved. Why non-human accounts end up with broad rights, how Dropbox Sign contained it, and a least-privilege checklist for the service accounts a small SaaS runs.
- CircleCI 2023: a session cookie from an engineer's laptop, and every customer told to rotate every secret
On 4 January 2023 CircleCI told every customer to rotate every secret they had stored there. Malware on one engineer's laptop had taken a single sign-on session that had already passed 2FA; the intruder used it to reach production, took customers' environment variables and keys, and the keys that decrypted them. What a CI provider holds for you, how the breach went step by step, and how to make your own rotate-everything day hours instead of weeks.
- SolarWinds 2020: the build server that added a backdoor to a signed update
In 2020 a small program on SolarWinds' build server swapped one source file while Orion was being compiled, so the signed update carried the SUNBURST backdoor to about 18,000 customers. The source code in the repository stayed clean. How SUNSPOT worked, how SUNBURST waited two weeks and hid in normal traffic, how FireEye found it from one extra MFA phone, and what build integrity means for a small software shop.
- 0ktapus 2022: text-message phishing that collected Okta logins and codes at Twilio, and why Cloudflare's security keys held
In summer 2022 the same text message reached staff at more than 130 companies and sent them to a copy of their Okta sign-in page. At Twilio, staff typed passwords and one-time codes and the attacker reached 209 customers, Signal and Okta users among them. At Cloudflare, three staff typed their password too, and hardware security keys refused the fake domain. Step by step, and the cheap way a small team gets the same result.
- Most "AI crawlers" on our server were fake: 14 days of nginx logs checked against the published IP lists
63,603 requests called themselves GPTBot, ClaudeBot, PerplexityBot, Googlebot and friends over 14 days. 80% came from addresses the companies don't use, and most asked for .env files, .git/config and Terraform state. The numbers per name, what the fakes wanted, and how to check your own logs.
- Storm-0558 2023: one stolen signing key, forged sign-in tokens for 25 organisations' mail
In 2023 a China-based group signed its own Microsoft sign-in tokens with a consumer key from 2016, and Exchange Online accepted them for government and enterprise mailboxes. No password was guessed and no MFA prompt appeared. It was caught by the US State Department, reading an audit log that most customers didn't have because it cost extra. What a signing key is, why 'which key may sign for whom' matters, and which logs a small tenant should keep.
- Kaseya VSA 2021: one IT management tool, about 1,500 businesses encrypted in a weekend
On Friday 2 July 2021, REvil used unpatched flaws in Kaseya's on-premises VSA servers to send a fake 'hot-fix' to every PC those servers managed. About 60 IT providers and up to 1,500 of their small-business clients were encrypted before the holiday weekend began. How the push worked step by step, why the remote-management agent was the perfect delivery system, and the questions a small firm should ask the IT provider that holds admin on every machine.
- CitrixBleed 2 (2025): the same kind of memory leak two years later, and what changed in the response
In June 2025 a NetScaler gateway login request with one bare field, 'login' with no value, returned bytes of the device's memory, and with them other users' session tokens. It was the 2023 Citrix Bleed story again. Attacks began within days of the patch, the vendor said for weeks it had seen none, and the session-kill commands it published missed most session types. A side-by-side of 2023 and 2025, and the runbook a gateway owner needs before the third one.
- Is your own site showing visitors a fake CAPTCHA? What a ClickFix injection looks like from the owner's side
Thousands of hacked WordPress sites now greet visitors with a fake 'verify you are human' box that copies a PowerShell command and asks them to press Win+R. The owner usually never sees it. How the script gets in, the three signs osec Scan's clipboard-lure check looks for, what a finding and a 'have a look' mean, and the clean-up steps.
- Citrix Bleed 2023: session tokens leaked from the gateway, and why patching didn't log attackers out
CVE-2023-4966 let anyone ask a NetScaler gateway for its OpenID settings with an oversized Host header and get back memory holding live session cookies: no password, no MFA. LockBit affiliates used it against Boeing's distribution arm, and 35.9 million Xfinity customers were exposed. Why patched boxes stayed compromised until sessions were killed, and what that means for any login gateway, including your app's.
- Codecov 2021: a modified uploader script that read CI secrets for two months
From 31 January to 1 April 2021 an intruder kept one extra line in Codecov's Bash Uploader. Every CI build that piped the script into a shell sent its git remote and all its environment variables out; HashiCorp had to rotate the key that signs its downloads. How a key leaked from a Docker image, why nobody noticed for two months, and the six pipeline habits that break each step.
- Salesloft Drift 2025: OAuth tokens from a chatbot integration opened hundreds of Salesforce orgs
In August 2025 an intruder used OAuth tokens taken from Drift, a website chat tool owned by Salesloft, to query the Salesforce data of hundreds of its customers, then searched the support cases for AWS keys, Snowflake tokens and passwords. No customer's password was guessed and no MFA prompt appeared anywhere. Why an integration's token is a sign-in without a second factor, and how to list, scope and revoke the ones your own accounts have handed out.
- Cloudflare, Thanksgiving 2023: four credentials left unrotated after the Okta breach
After Okta's support system was breached in October 2023, Cloudflare rotated the credentials that had leaked. It missed one service token and three service accounts it believed were unused. A month later those four opened its internal wiki, bug tracker and source code server. The day-by-day account from Cloudflare's own report, why "we think that one is dead" is the usual way rotation fails, and a rotation checklist that starts from an inventory and not from memory.
- 3CX 2023: a supply chain attack that began with another supply chain attack
A 3CX employee installed a trading program on a personal computer in 2022. Its installer had been tampered with at the vendor. A year later 3CX's own signed desktop phone app was shipping a backdoor to customers, and for a week antivirus alerts about it were discussed as false positives. How one poisoned installer led to the next, why a valid signature ended the argument too early, and what a small firm can check about the software it installs for staff.
- Shai-Hulud 2025: the npm worm that republished itself with stolen maintainer tokens
In September 2025 a script that ran during npm install searched developers' machines with a secret scanner, pushed what it found to public GitHub repositories, then used the npm tokens it had just stolen to publish infected versions of the victim's other packages. More than 500 packages were removed; a second wave in November reached about 700 more. How each step worked, and what trusted publishing, blocked install scripts and short-lived tokens change.
- CosmicSting 2024: how one Magento bug let skimmers into thousands of stores
In June 2024 a flaw in Magento and Adobe Commerce let anyone read one file from a store's server. That file holds the store's encryption key, and the key signs admin API tokens. Within months 4,275 stores had card skimmers on their checkout, many of them fully patched. Step by step, and why patching without replacing the key left the door open.
- Ivanti Connect Secure 2024: two chained zero-days on the VPN gateway
In December 2023 someone began running commands as root on Ivanti VPN appliances without signing in. For three weeks after the flaws were made public there was no patch, the vendor's own integrity checker could be told to skip the changed files, and US federal agencies were ordered to unplug the product. Step by step, and what an organisation with one VPN appliance should do the next time.
- tj-actions/changed-files 2025: the GitHub Action that printed secrets into public build logs
In March 2025 every version tag of an action used by 23,000 repositories was pointed at code that read the build runner's memory and printed its secrets into the log. The chain started four months earlier with one token in another project. The timeline, why 218 repositories leaked and not 23,000, and the five settings that would have stopped each step.
- Lumma Stealer: a $250-a-month infostealer, fake CAPTCHAs and the takedown that lasted weeks
Lumma was rented to about 400 criminals at a time and found on 394,000 Windows PCs in two months. How the business worked, how a fake "I'm not a robot" box gets a person to run the malware themselves, what one stolen log holds, what the May 2025 takedown changed, and what a site owner can do about sessions stolen from customers' PCs.
- LastPass 2022: from a developer laptop to encrypted vault backups
Two linked break-ins: source code from a developer's laptop in August 2022, then a keylogger on a senior engineer's home PC, planted through a media server that hadn't been updated since 2020. The prize was a backup of customer vaults. What the encryption protected, what it left readable, and why old accounts were cracked first.
- Change Healthcare 2024: one Citrix portal without MFA, a $22 million ransom
On 12 February 2024 someone signed in to a Change Healthcare remote-access portal with a stolen username and password. The portal had no second factor. Nine days later ransomware stopped a large part of US medical billing, the company paid $22 million, and a second gang asked for more. Step by step, and the controls that break each step.
- XZ Utils backdoor: a two-year social-engineering supply chain attack
A friendly contributor spent two and a half years earning maintainer rights on xz, helped by invented critics who pressured its tired volunteer maintainer, then hid a key-gated SSH backdoor in test files and a tarball-only build script. It was caught by a half-second slowdown. Step by step, and the controls that break each step.
- Session cookie theft after MFA: why passkeys don't fully fix it
Passkeys stop phishing and password reuse, but the session cookie a site issues after sign-in is still a bearer token, and infostealers copy those by the billion. How the theft works step by step, what Chrome's Device Bound Session Credentials change, and what a small site can do this month.
- Okta 2023: how HAR files in a support system became hijacked admin sessions
A support-system password saved in a personal Chrome profile let an attacker read customer support files at Okta for three weeks. Some were HAR files holding live session cookies, which were replayed against 1Password, BeyondTrust and Cloudflare. Step by step, why logging missed it, and the controls that break each step.
- Alternative to Cloudflare Access for public sites: passwordless sign-in with osec Auth and Login Protection
Cloudflare Access is free for up to 50 users and puts a login in front of your apps. Five reasons a small site with its own sign-in can use osec Auth and Login Protection instead: no proxy, no OAuth setup, email codes, Google and Apple. Includes pricing and where Cloudflare Access is the better choice.
- Alternative to Cloudflare Bot Management: osec Bot Gate and Fingerprint Lite for small sites
Cloudflare's bot protection starts free with Bot Fight Mode, then $20 and $200 a month, or a sales quote. Five reasons a small site can use osec Bot Gate, Fingerprint Lite and Captcha instead, with a proxy-free setup, pricing and where Cloudflare is the better choice.
- Alternative to Cloudflare Turnstile: five reasons to use osec Captcha instead, and when not to
Turnstile is free with no request cap, so it's the default choice for most small sites. Five reasons to use osec Captcha, Signup Checks and Bot Gate instead: no Cloudflare account, a check you run yourself, and a free tier of 500 a day. Includes pricing and where Turnstile is better.
- Alternative to reCAPTCHA: five reasons a small site can switch, and what it costs
reCAPTCHA's free tier is 10,000 assessments a month, then billing starts. Five reasons to try osec Captcha, Login Protection and Signup Checks instead, with pricing at each level and where reCAPTCHA is still the better choice.
- Fingerprint Pro vs osec Fingerprint Lite: what visitor identification costs, and what you give up
Fingerprint Pro costs $99 a month for 20,000 identifications, then $4 per 1,000. osec Fingerprint Lite gives 500 a day free and charges by capacity. Here's the price at each volume, and where Lite is weaker.
- An nginx hardening checklist: timeouts, body limits and a check script
Slow clients, oversized bodies and version banners are cheap to fix in nginx. Set the limits once, then run a bash script that checks them from outside, so drift shows up.
- Block whole networks cheaply with ipset and a bash script
One firewall rule per address doesn't scale. ipset keeps thousands of addresses and ranges in one hash set, so a single iptables rule drops all of them at a fast lookup.
- Security patches on their own: unattended-upgrades on Ubuntu
Most breaches ride on bugs that already have a patch. unattended-upgrades installs security updates each day, leaves reboots to you, and writes a log you can check with a short script.
- Close probe paths and bad user agents in nginx with return 444
Most bot traffic asks for the same few paths. nginx can close those connections with the non-standard code 444 before your app wakes up, with no response for the scanner to read.
- A Python report of bot pings, from your own nginx access log
Before you write a single rule, measure. This script reads the nginx access log, groups requests by IP, counts probes, and prints the top offenders you can ban or rate limit.
- fail2ban for SSH, then a recidive jail for repeat offenders
A ten-minute ban just makes a bot wait. The recidive jail reads fail2ban's own log and bans repeat offenders for a week, across every jail.
- A default-deny firewall for a small server with ufw
Open only the ports you serve, rate-limit SSH, and keep the rules in a script you can rerun. ufw covers it in about twenty lines, with a rollback path.
- SSH hardening: keys only, one allowed user, and a way back in
Password guessing against port 22 is constant background noise. Turn off password logins, limit who can log in, and test from a second session before you close the first.
- nginx rate limiting that slows bots without locking out people
limit_req counts requests per key over time. Set it per IP for login and API paths, allow short bursts for real browsers, and return 429 so you can see what happened.
- fail2ban for nginx: ban the bots that keep knocking on /.env and /wp-login.php
Scanners request /wp-login.php, /.env and /phpmyadmin hundreds of times a day. fail2ban reads your nginx logs and bans the IPs that do it, with two jails you can test in five minutes.
- Sign-in rules that count emails and wrong codes, not just IPs
Credential stuffing and code-bombing hit sign-in pages first. A generic rate limit counts per IP over seconds; attackers rotate IPs over hours. osec Login Protection adds per-email and per-IP code limits per hour, wrong-code limits, IP lists, allowed email domains and disposable-email blocking to osec Auth.
- AI scrapers hammering a small site? A bot gate without moving your DNS
AI crawlers now make up a large share of traffic to small sites, and many ignore robots.txt. osec Bot Gate is a two-line middleware for Node, Python or PHP: new visitors solve one invisible check, then browse for a week. Search engines pass. Nothing is proxied through us.
- We went from F to A+ on security headers with one nginx file. Grade your site
Our own site scored F: no HSTS, no Content-Security-Policy, no clickjacking protection, and a server version in every response. One shared nginx snippet took it to A+. Here's the list, and a free scanner that grades your site with copy-paste fixes.
- Certificates now expire in 200 days, soon 47. Catch it before your visitors do
Public TLS certificates dropped to a 200-day maximum in March 2026, 100 days next year and 47 days in 2029. More renewals mean more chances to break. osec Monitor watches uptime, certificate expiry and domain expiry, and emails you first.
- reCAPTCHA's free tier is now 10,000 checks a month. Here's a private, invisible swap
Google's reCAPTCHA now bills past 10,000 assessments a month. What that means for contact forms, signups and logins, and how osec Captcha replaces it with an invisible proof-of-work check: no image puzzles, no tracking, reCAPTCHA-style siteverify.
- Passwordless sign-in for any site in minutes: email codes, Google and Apple with osec.one (free)
The no-paragraph guide: what osec.one gives your site, why there's no Google or Apple OAuth setup on your side, and the exact steps from new project to working login. For existing sites, new sites, SaaS apps and larger teams. Free.
- 23andMe 2023: 14,000 reused passwords, 6.9 million people exposed
An attacker logged in to about 14,000 23andMe accounts with passwords leaked from other sites, then used a data-sharing feature to scrape profiles of 6.9 million people. How the credential stuffing campaign worked, the warning signs that were missed, and what the UK regulator said should have been in place.
- British Airways 2018: 22 lines of JavaScript and a £20 million fine
In 2018 attackers used a supplier's login to get into British Airways' network, then added 22 lines to a JavaScript library on the payment page. For two weeks card details went to a look-alike domain. A walkthrough of the Magecart skimming chain, how it was found, and the controls that would have broken it.
- MGM and Caesars 2023: when the help desk resets MFA for the attacker
In 2023 Scattered Spider skipped password cracking and phoned IT support instead, asking for passwords and MFA to be reset on privileged accounts. MGM put the cost at about $100 million. How help-desk social engineering works step by step, and the verification and identity controls that stop it.
- Uber 2022: how push-notification fatigue turned one bought password into admin access
In September 2022 an attacker with a contractor's stolen password kept sending MFA push requests until one was approved, then found admin credentials in a script on a network share. A step-by-step look at MFA fatigue, why plain push approval is weak, and the controls that break each step.
- npm 12 and the 2026 GitHub Actions hardening: what changes in your pipeline
After a run of worm-style npm compromises and mass GitHub Actions backdoors, GitHub shipped its biggest set of supply-chain defaults yet: install scripts off in npm 12, staged publishing, safer pull_request_target, read-only caches and a network firewall for runners. What each change does and what to update this week.
- Tycoon2FA: how a phishing kit walked past MFA at 500,000 organisations a month
Tycoon2FA sold adversary-in-the-middle phishing as a subscription. It relayed real logins to Microsoft and Google, let victims complete their MFA, then kept the session cookie. A breakdown of the kit, why codes and push prompts didn't help, and what does.
- Polyfill.io: when a trusted script tag turned against 380,000 sites
Hundreds of thousands of websites loaded JavaScript from cdn.polyfill.io. After the domain changed hands, it started serving code that redirected some mobile visitors to scam sites, and tried hard not to be seen. How the attack worked and how SRI, CSP and self-hosting would have contained it.
- Snowflake 2024: how years-old infostealer logs opened 165 data warehouses
The 2024 Snowflake customer breaches weren't a Snowflake vulnerability. Attackers logged in with passwords stolen by infostealer malware, some as old as 2020, and used plain SQL to stage and download whole tables. A walkthrough of the method and the three gaps it relied on.
- How a quiet password spray reached Microsoft's leadership inboxes
Midnight Blizzard didn't use an exploit to get into Microsoft's corporate email. It guessed one password on a forgotten test account, then chained OAuth apps until it could read mailboxes. Here is the chain, step by step, and the controls that break each link.