Lead · Supply chain

Somebody A/B tested malware on 178 organisations, and the websites helped

Malicious JavaScript served through Optimizely's experimentation platform reached visitors of CNN, the BBC, Zoom, Salesforce and GitLab, with a card skimmer down one path and an iOS exploit kit down another.

confidence 0.92

On Monday 16 November an attacker got into Optimizely Web Experimentation, which is the service a great many large sites use to change what a page does without shipping a release, and used it to serve malicious JavaScript to visitors of at least 178 organisations, among them CNN, the BBC, Fox News, Zoom, Salesforce and GitLab; one branch of that script was a payment-card skimmer, another was an iOS exploit kit, and Optimizely has since confirmed that an attacker exploited a flaw in the platform. The sites did exactly what they were built to do. They loaded whatever the tag said.

Censys sinkholed one of the kit's fallback domains and, in two and a half hours, logged 509 victim hostnames and 13,808 victim requests, with victims running iOS 17.4 through 26.7 and arriving from major news, transportation and shopping brands. A follow-on investigation traced malicious scripts delivered through four web services rather than one: Optimizely, Kissmetrics, StorifyMe and minute-ly. So the question this morning is not whether you use Optimizely. It is how many strangers currently hold a pen over your checkout page.

There is no clever mitigation to recommend that you did not already know about. Content Security Policy, subresource integrity where the vendor's model allows it, a real inventory of third-party script, and a conversation with whoever owns the tag manager that does not end with the sentence 'that's a marketing system'.

INST-203.

Why it mattersIf you cannot name, this morning, every third party permitted to inject script into your payment pages, that list is today's incident scope.

Today’s revelation

Every tag you added for marketing is a remote code execution you have agreed to renew annually.

— Sturgeon’s Revelations · Probably Fine Daily No. 52

§2

Top stories

The rest of the last 24 hours, with receipts.

Vulnerabilities

One new KEV entry, and it is the machine that tells your other machines what to be

CISA added CVE-2026-115731, a deserialisation flaw in Perforce Puppet Enterprise's report-processing code: a client holding any CA-signed agent certificate can execute arbitrary code on the Puppet Server as the service user. Its EPSS score is 0.00366, which is a statement about the world's past behaviour rather than about yours. Separately, a CTI channel flagged Cisco BroadWorks CVE-2026-20320, an XML entity reference issue exposing sensitive information, as under active exploitation with mitigations due by 24 November, and a high-severity UnrealIRCd flaw, CVE-2026-114713, is reported exploited in attacks.

Why it mattersEvery host you ever enrolled with a signed Puppet agent certificate is now part of the exposure list for the server that configures the estate.

confidence 0.85

Vulnerabilities

Java4Shell got a nickname before anyone agreed on the number

Oracle shipped fixes for 11 Java SE flaws on 17 November, all of them potentially exploitable remotely without authentication, and attention promptly collapsed onto CVE-2026-114938 in the Compiler component of Java SE and GraalVM. BleepingComputer described it as allowing unauthenticated remote code execution against Jenkins, Atlassian Jira, Confluence and Keycloak; The Hacker News reported a CVSS 9.2 figure as unconfirmed on Tuesday and published exploit details on Wednesday, saying independent labs confirm memory corruption and data leaks. Same bug, two registers, one logo.

Why it mattersPatch the runtimes you have rather than wait for the severity argument to resolve; the applications named are the ones holding your build pipelines and your single sign-on.

confidence 0.60

Threat actors

The remote access tool is not the vector this week, it is the campaign

CVE-2026-84869 in ConnectWise ScreenConnect, CVSS 9.9, lets a malicious guest in a remote session push files to a technician's workstation and execute them with elevated privileges; more than 40 servers received over 688,000 requests exploiting it, and between 3 and 13 November attackers enrolled over 2,200 fake guest sessions, disguised with crypto exchange names, across 118 ScreenConnect servers belonging to 109 MSPs and organisations. ConnectWise says the technician client must be on 26.6.5 or later, and a toolkit tracked as SCKit has been abusing cloud-hosted instances since 11 November. In the same window N-able patched 17 more N-central flaws while attackers exploited earlier ones to deploy crypto miners, and CVE-2026-90436 in Apache Axis was seen exploited against Atlassian Confluence Data Center and N-central.

Why it mattersIf a managed service provider can reach your endpoints, their support console is an unlogged administrative path into you, and this week it is being enumerated at scale.

confidence 0.85

Ransomware

A leak site nobody had heard of posted five large US names in about two minutes

Between 08:14 and 08:16 UTC on 17 November, a group calling itself Inception claimed Advanced Micro Devices (57 GB, 185,391 files, including an internal hostname inventory of 1,665 endpoints and 58,678 employee identities), Cigna Healthcare (9 GB), Henry Schein, Paychex (68 GB, 123,000 files, including SSN and bank-account workbooks for roughly 300 people) and PepsiCo (over 500,000 employee records). The descriptions read like incident and change records: close notes, work notes, access-management systems named by product. All of it is unverified claim, and all of it is the same shape.

Why it mattersIf your service-desk or ITSM tenancy is shared with a partner, ask the vendor today rather than waiting for a victim list to be published for you.

confidence 0.60

Ransomware

331 claims in seven days, 48 groups out of 278, and three of them running hot

Our tempo count has 331 leak-site claims in the last seven days from 48 of the 278 groups tracked, with incransom at 56 against 7.4 expected from its own prior rate, n0n at 74 against 36.91, and netrunner at 4 against 0.24. Hackmanac's week totals 497 attacks across 65 countries, with n0n the most active actor at 72 and the United States taking 31.4 per cent of incidents at 156. The Gentlemen posted 58 items on 17 November alone, including DIRECTV Latin America and AARP, while Vortex's healthcare claims carry record counts and hard dates: Owens-Minor at 57,985,343 records and Caris Life Sciences at 21,432,021, with deadlines of 19 and 20 November. A Dragos summary doing the rounds has industrial ransomware up 11 per cent in Q3 2026, manufacturers 69 per cent of victims, Qilin on 195 industrial victims and The Gentlemen on 167.

Why it mattersTwo Vortex deadlines land inside 48 hours; if you are in the healthcare or pharma column, the communications draft is today's work, not Friday's.

confidence 0.80

AI & security

Machine speed arrives, still riding a key somebody forgot to turn off

Google says adversarial use of AI has moved past prompting into autonomous multi-agent frameworks — ARTEX, CAIRN, CyberStrikeAI and PentAGI — used to run multi-stage intrusions, with China-nexus actors reported breaching 10 organisations using Cairn. Barracuda ties its November cloud breach to ARTEX plus a cloud access key that was never retired after a third-party AI vendor's project, with around 10 cloud accounts touched in 44 hours; a separate CyberStrikeAI intrusion made 56 AWS API calls in 141 milliseconds across four accounts using eight compromised identities. On the same day, Project Zero announced a move to a 60-day disclosure window in February 2027 and 30 days in January 2028, citing AI-accelerated discovery and exploitation.

Why it mattersRate-based detections and human-paced approval gates were tuned for an attacker who types; budget for the one that does not.

confidence 0.75

AI & security

The AI provider key is now initial access, and it is being collected on an industrial schedule

Cursor traced the malicious Claude Code extension copies to a key stolen from AI infrastructure provider Fireworks, and Uber is notifying nearly 150,000 drivers that names, addresses and driver's licence numbers were taken after that same provider was breached in early October. Amazon's decoy servers captured AI agents working through internet-exposed AI applications — LiteLLM, One API, Dify — specifically hunting the provider keys stored inside, and SANS honeypots have been logging requests chaining CVE-2026-42271 and CVE-2026-48710 for remote code execution. Meanwhile, credentials exposed in public Docker Hub images are being extracted and used within as little as eight seconds of publication.

Why it mattersPut AI provider keys in the same rotation, scanning and revocation tier as cloud credentials, because the people collecting them already have.

confidence 0.80

Oddities

The cold chain has six unpatched flaws and no login requirement

Midnight Blue reported six flaws in Danfoss AK-SM800A refrigeration controllers that allow a remote attacker to take over a controller without logging in, with no patch available and the newest firmware, version 5.4.0, still vulnerable. In the same window, VulnCheck disclosed 14 vulnerabilities in Copeland XWEB EVO controllers, two of which attackers began probing quickly; CVE-2026-124015 lets unauthenticated attackers set the password of any account flagged to reset at next login, a public exploit is circulating, and the product line reached end of service in March 2026 with no fixed version listed.

Why it mattersThese sit in supermarket, food production and pharmaceutical estates and are usually procured as building services, which is why they are absent from the vulnerability management scope you presented in October.

confidence 0.85
§3

Hidden

The daily deep read: what is under the feeds, not on top of them.

Under the feeds: 396 of 858 items landed on a single day, and a good slice of that velocity belongs to the scrapers rather than the attackers.

  1. Exception 01

    Inception's five victims look like one upstream compromise, not five intrusions

    AMD, Cigna, Henry Schein, Paychex and PepsiCo were posted within about two minutes with ITSM/ServiceNow-shaped data, across five sources, on a day when the median Inception count on other days is zero. That pattern fits one shared SaaS or third-party compromise better than five separate break-ins.

    confidence 0.60Speculative
    Wrong if
    Victims confirm separate, unrelated intrusion paths, or the data proves to be old or recycled.
    Watch
    Vendor or ServiceNow-partner disclosures and victim 8-Ks over the next one to three days.
  2. Exception 02

    The Gentlemen's 58-item day is largely a batch dump

    Roughly 17 victims across about 12 countries appeared inside five minutes, and the Atlas stories share one timestamp, so the apparent velocity is a disclosure or scrape batch and per-victim timing should not be trusted.

    confidence 0.55Speculative
    Wrong if
    Source timestamps turn out to spread over many hours, or new Gentlemen victims keep appearing at a similar pace over the following days.
    Watch
    Gentlemen post rate over the next three days, and whether DIRECTV Latin America confirms.
  3. Exception 03

    Java4Shell severity is contradicted across sources

    One set of reports presents pre-authentication RCE against Jenkins, Jira, Confluence and Keycloak as established; another carries a CVSS 9.2 figure explicitly marked unconfirmed. Treat the nickname as marketing until a reproducible proof of concept against default installs is public.

    confidence 0.60
    Wrong if
    A working public PoC demonstrates pre-auth RCE against default installs of those products.
    Watch
    Oracle advisory updates and statements from Jenkins and Atlassian.
  4. Exception 04

    NVD went quiet after 03:31 on 17 November

    All 77 NVD items in the window fall before that timestamp, so the later CVE picture leans on news and Telegram. This may be an ingest gap rather than real quiet, and it caps confidence on anything published late.

    confidence 0.50Speculative
    Wrong if
    NVD publishes normally through the rest of the window and the items were simply not ingested.
    Watch
    Whether NVD ingest resumes in the next window.

Nothing to seeFugaku was taken offline after a breach, NSANet suffered unspecified disruption, Apple reported iCloud Mail and sign-in outages and Tor shipped an emergency update against an ongoing DDoS, all inside 24 hours, and nothing in the items connects any of them to each other; no cause is given for the NSANet or iCloud disruptions. The Lighter exchange exploit, the Telegram widget RCE demos and the Batflat proof of concept each rest on single or unverified sources. The feed-to-feed lead-lag figures are mostly scraper cadence, not anybody knowing first.

Deep read: 859 items read

§4

Forecast

Dated, falsifiable calls. We mark our own homework in public.

Today’s calls

  1. DueWed 25 Nov7d

    incransom posts at least 20 further leak-site claims in the seven days to 25 November, continuing the surge our tempo count flagged.

    Wrong ifFewer than 20 incransom claims appear across ransomware.live, ransomlook and Atlas in that window.

    f-2026-11-18-1

    Confidence0.70
  2. DueMon 30 Nov12d

    CVE-2026-114938, the OpenJDK flaw being marketed as Java4Shell, is not added to CISA KEV by 30 November.

    Wrong ifCISA adds CVE-2026-114938 to the Known Exploited Vulnerabilities catalogue on or before 30 November, which makes this call wrong.

    f-2026-11-18-2

    Confidence0.30
  3. DueWed 25 Nov7d

    The Gentlemen post 19 or fewer new leak-site claims in the seven days to 25 November, because 17 November was a batch rather than a rate.

    Wrong ifTwenty or more new Gentlemen victims appear across the leak-site trackers in that window.

    f-2026-11-18-3

    Confidence0.60

Scorecard

2 right · 1 wrong

  1. wrong

    PixelLeaks does not repeat yesterday's volume: fewer than 10 new claims in the seven days to 23 November.

    Called Mon 16 Nov · due Mon 23 Nov

    darkwebsonar reports 18 PixelLeaks listings in the past week, and two new PixelLeaks claims appear in ransomlook on 2026-11-18, confirming ≥10 claims in the window; falsification condition met.

  2. right

    N0n posts at least 10 further leak-site claims in the seven days to 24 November, rather than going quiet after the bulk dump.

    Called Tue 17 Nov · due Tue 24 Nov

    Ninja Signal tempo shows N0n surging with 74 claims in 7 days, well above the ≥10 threshold; ransomlook also shows multiple N0n leaks posted this period.

  3. right

    incransom's surge holds: at least 15 claims in the seven days to 24 November.

    Called Tue 17 Nov · due Tue 24 Nov

    Ninja Signal tempo shows incransom surging with 56 claims in 7 days, well above the ≥15 threshold.

Glossary of terms you will be expected to nod at

Experimentation platform (n.)
A third party with write access to what your website says, chosen by marketing, tested on customers, inventoried by nobody, and absent from the asset register on the grounds that it is not an asset, it is a tag.
§5

From our edge

Live attacks on our own servers in the last 24 hours. Yes, really ours.

Our own edge, last 24 hours: 135 hostile sources profiled — 70 scanners, 58 bot crawlers, 7 targeted operators — sending 21,949 requests between them, of which 81 stopped looking and started trying. Twenty clusters of addresses shared a behavioural fingerprint, one of which, DNA dfef4501b216, spans 22 separate IPs, an arrangement any vendor would describe as a mesh. One scanner fired 268 requests at /.env, /backend/.env, /api/.env and /src/.env in about 0.9 seconds; another sustained 192 requests a minute. The highest hostility on the board, 7.7 out of 10, belonged to 85.204.70.100, which sent 20 requests for wlwmanifest.xml and xmlrpc.php under /login and reached credential access, and 129.226.144.66 was still spraying php-cgi strings shaped like CVE-2024-4577, because somewhere a box still answers. In total 10,080 addresses from 124 countries, 6,761 of them seen here for the first time today. The raw logs never left the server.

135hostile sources profiled
21,949requests from them
81reached exploitation
20shared-DNA clusters
7.7/10top hostility
10,080addresses at our edge
6,761first seen today
124countries

The WTF list

  1. WTF #1

    The most determined visitor

    85.204.70.100, a targeted operator (hostility 7.7/10), sent 20 requests to one of our sites, probing for /login/wp-includes/wlwmanifest.xml, /login/xmlrpc.php, /login/blog/wp-includes/wlwmanifest.xml, /login/web/wp-includes/wlwmanifest.xml.

    Why it’s weirdHighest hostility on the board; its kill chain reached credential-access.

    attribution: unknownTue 17 Nov, 11:00 UTC

    Detail
    kill chain: reconnaissance → weaponization → exploitation → credential-access · DNA 82a939927b85
  2. WTF #2

    Same DNA, different addresses

    20.48.89.150, a scanner (hostility 6.7/10), sent 334 requests at 192 a minute to one of our sites, probing for /classwithtostring.php, /admin/controller/extension/extension/ultra.php, /lock360.php, /flower.php.

    Why it’s weird22 IPs share one behavioural fingerprint (DNA dfef4501b216): one tool or one operator rotating addresses: 20.28.214.139, 85.204.70.94, 85.204.70.88, 20.48.89.150, 40.123.240.128 and more.

    attribution: unknownTue 17 Nov, 23:00 UTC

    Detail
    kill chain: reconnaissance → weaponization · DNA dfef4501b216
  3. WTF #3

    Trying CVE-2024-4577 on us

    1 address sent 1 requests shaped like CVE-2024-4577 exploitation (/api/threat-check/php-cgi%20query%20string%20parameter%20vulnerability, /threat-check/php-cgi%20query%20string%20paramet), including 129.226.144.66.

    Why it’s weirdA known, named exploit still being sprayed at the internet; our edge saw it today, whatever the patch notes say about the year.

    attribution: unknown

    Detail
    probe family: php-cgi
  4. WTF #4

    The strangest shopping list

    20.203.159.118, a targeted operator (hostility 6.2/10), sent 209 requests at 2 a minute to 2 of our sites, probing for /wp-content/plugins/hellopress/wp_filemanager.php, /this_is_a_new_hello_world.php, /king.php, /rich.php.

    Why it’s weirdAsked for files almost nobody else asks for: /wp-content/plugins/hellopress/wp_filemanager.php, /this_is_a_new_hello_world.php, /king.php, /rich.php, /wp-optimizer.php.

    attribution: unknownTue 17 Nov, 23:00 UTC

    Detail
    kill chain: reconnaissance → weaponization → exploitation · DNA 938595d2ae44
  5. WTF #5

    In a hurry

    34.156.155.78, a scanner (hostility 5.0/10), sent 268 requests to one of our sites, probing for /.env, /backend/.env, /api/.env, /src/.env.

    Why it’s weirdFired 268 requests in about 0.9 seconds.

    attribution: unknownWed 18 Nov, 05:00 UTC

    Detail
    kill chain: reconnaissance → weaponization → exploitation · DNA ce6ecdfacbe3
  6. WTF #6

    Somebody we might know

    103.59.161.225, a scanner (hostility 5.9/10), sent 39 requests to one of our sites, probing for /wp-admin/, /wp-admin/css/, /wp-admin/images/, /wp-admin/includes/.

    Why it’s weirdThe sensor's most confident campaign match of the day: WordPress automated vulnerability scanner / mass-scan operation at 55%.

    attribution: WordPress automated vulnerability scanner / mass-scan operation (sensor's best match, 55%)Wed 18 Nov, 05:00 UTC

    Detail
    kill chain: reconnaissance → weaponization · DNA 18e39c4b41fb
§6

Tempo

Who is posting faster than their own normal. Computed from the Signal graph, not written by the editor.

331claims, 7 days
48of 278 groups active
N0n74 this week · surging
incransom56 this week · surging
thegentlemen39 this week · surging
  • Surgingincransom — 56 claims in 7 days against 7.4 expected from its own prior rate (p < 0.001)
  • SurgingN0n — 74 claims in 7 days against 36.9 expected from its own prior rate (p < 0.001)
  • Surgingnetrunner — 4 claims in 7 days against 0.2 expected from its own prior rate (p < 0.001)
  • SurgingShadowByt3$ — 5 claims in 7 days against 0.7 expected from its own prior rate (p < 0.001)
  • SurgingEndZone — 4 claims in 7 days against 0.8 expected from its own prior rate (p = 0.008)
  • Surgingthegentlemen — 39 claims in 7 days against 26.6 expected from its own prior rate (p = 0.014)

Gone quiet: killsec (61 days), securotrop (61 days), global (67 days), embargo (70 days).

Sector pressure, 30 days

  • UpRetail & E-Commerce — 1.4× its own prior share (105 claims, 78 expected)
  • UpHealthcare — 1.2× its own prior share (137 claims, 114 expected)
  • DownAgriculture and Food Production — 0.4× its own prior share (18 claims, 42 expected)
  • DownTransportation — 0.6× its own prior share (34 claims, 54 expected)

26% of claims carry no sector label.

Brand churn

Of the 240 leak-site brands first seen since 2021-09-30, 75% were still claiming 30 days after their first post and 70% were still claiming 60 days after their first post and 67% were still claiming 90 days after their first post and 61% were still claiming 120 days after their first post (Kaplan-Meier; quiet = 60 days without a claim). 18% of the 217 old enough to judge posted on one day and never again; 37% made at least half their claims in their first week. New brands this month: 9 — newest: Inception (2026-11-17, 5 claims), Curious Worm (2026-11-11, 17 claims), osiris (2026-11-11, 8 claims).

Brands born before coverage began are excluded (their true age is unknown); a rebrand counts as one brand going quiet and another being born.

Targeting

Over the last 180 days, among brands with at least 20 labelled claims: safepay put 33% of its claims on Germany (6.3× every other brand's share); N0n put 55% of its claims on the United States (1.4× every other brand's share); SilentRansomGroup put 100% of its claims on Professional Services (6.4× every other brand's share). Widest spread: nightspire across 10.7 effective sectors.

Lift = the brand's share of a sector or country over every other brand's share in the same window; exact binomial tails, Bonferroni-adjusted for the categories tested. Claims with no sector or an Unknown/Other label are excluded.

Posting clock

Over the last 180 days, among brands with at least 30 claims: N0n released 50% of its claims on Mondays; clop released 94% of its claims on Wednesdays, half of them in 13:00–15:59 UTC; akira posted only 2 of 196 claims at weekends and half its claims in 12:00–13:59 UTC; half of thegentlemen's claims appeared in 06:00–09:59 UTC. Leak sites overall are 36% quieter per weekend day than per weekday.

all brands by hour of day, 00–23 UTC; half of all claims land in 09:00–17:59 UTC.

Times are when trackers first saw each claim on the leak site, in UTC: not the attack time, not the operator's clock, and a site that is only reachable at certain hours shows the same pattern; batch releases show as spikes. Hour-of-day concentration: Rayleigh test; weekday shares: exact binomial against every other brand, Bonferroni-adjusted.

Leak-site claims as discovered by trackers, not confirmed breaches. Surprise compares the last 7 days with the group's own prior 173 days (Poisson). Victim claims: ransomware.live and leak-site trackers

Today’s numbers as JSON →

§7

Ride-alongs

Free, live tools. No sign-up, no sales call.

§8

Colophon

How this edition was made, plainly.

Edition 52 was made by machine and checked by code. Every item published by 15 public threat feeds in the 24 hours to 05:05 UK (859 items) was read in full: statistics computed in code, each slice read by Claude Sonnet, the structure synthesised by Claude Opus. Claude Opus then edited the paper. Every story cites the items it rests on; stories whose receipts did not resolve, or whose numbers were not in their sources, were dropped before publication (1 today). Forecasts are graded in public in later editions. 'From our edge' is live attack data from our own servers. Cost of this edition: $1.61. It is written by machine in the voice of ScottG, author of Everything Is Probably Fine, and nobody edited it by hand, which is either the point or the problem.

859items read
24hour window
$1.61cost of this edition
05:05 UTCgenerated

Sources this morning

  • ThreatFox193ok
  • Telegram125ok
  • Signal Atlas108ok
  • infosec.exchange88ok
  • Security news87ok
  • NVD77ok
  • ransomware.live61ok
  • RansomLook40ok
  • RSS26ok
  • atlas22ok
  • AlienVault OTX21ok
  • signal:ioc-family8ok
  • signal:kev1ok
  • CISA KEV1ok
  • Ransomware tempo · Ninja Signal graph1ok

ITEMS READ: 859SOURCES: 15STORIES DROPPED FOR LACK OF RECEIPTS: 1COST: $1.61STATUS: NOT ESPECIALLY FINE

OWNER: the machineEDITOR: the rulesCALLS OPEN: 3NEXT REVIEW: tomorrow 06:00 UK