Lead · Vulnerabilities

CISA added four actively exploited flaws on Friday. The deadline is tomorrow.

Apache Roller, VMware vCenter, SAP NetWeaver and Cisco Expressway went into the Known Exploited Vulnerabilities catalogue on 20 November with a remediation date of 23 November, which is Monday, which means the weekend was the plan.

confidence 0.93

Edition No. 56 · INST-207 · it is Sunday, and the deadline is tomorrow. The four are these: CVE-2026-129448 in Apache Roller, a deserialisation bug reachable over the network with no login and no click, CVSS 9.8, fixed in 6.1.6; CVE-2026-123890 in Broadcom VMware vCenter, an authentication bypass in the Directory Service; CVE-2025-42884 in SAP NetWeaver Enterprise Portal; and CVE-2026-76466 in Cisco Expressway, hard-coded credentials that let an unauthenticated remote attacker run commands as root on a clustered device.

The vCenter one has a history. Amazon-adjacent reporting and the OTX pulse both put CVE-2026-123890 in the hands of ARMORY CASTLE, the suspected PRC-nexus group formerly tracked as UNC5221, exploiting it as a zero-day from early September and, by one account, running for over two months with a custom tool called BONESAW that forged SAML tokens, which is the part that should make your identity team put the biscuit down. The Expressway bug went the other way: a proof-of-concept was published, and threat actors used it to breach a Russian organisation shortly afterwards, in the time it takes most of us to get a change ticket approved.

I have been the person reading a KEV entry on a Sunday with a mug going cold, and the honest thing to say is that the catalogue is not the problem and the patch is not the problem. The problem is that you do not know how many vCenters you have. Nobody does. The asset inventory is a photograph of a tide, taken in March, by a contractor who has since left.

Why it mattersFour unauthenticated-or-near-enough paths into identity, collaboration, ERP and virtualisation infrastructure, all confirmed exploited, all due Monday: find the boxes first, patch second, and check vCenter logs for token weirdness before you close the ticket.

Today’s revelation

An attack is called novel when the tool is new and the hole is old. The hole is always old.

— Sturgeon’s Revelations · Probably Fine Daily No. 56

§2

Top stories

The rest of the last 24 hours, with receipts.

AI & security

An AI agent took 148 actions in 16 hours and found one real zero-day. One is enough.

Amazon Threat Intelligence says Chinese-speaking operators pointed Cairn, an open-source autonomous penetration-testing framework, at two East Asian universities; across 148 agent actions over sixteen hours the only success was a previously unknown path-traversal flaw in MONAI Label, an open-source medical imaging server, which gave root on a university server and was then reused against the second university. The flaw is rated CVSS 9.8, lets an attacker with no account write files, and as reported there is no patched release. One account puts the time from pointing the thing at the target to a working exploit path at eleven minutes. In the same window, Bell Canada said an 'agentic' actor breached it on 17 November and took enterprise customer data including from a circuit-ordering system, the European Commission confirmed an AI-assisted attack that forced it to take services offline with no evidence of a breach, and a 41-module PowerShell toolkit with strong indicators of AI-assisted development turned up at an African stock exchange.

Why it mattersIf you run MONAI Label or anything like it on a research network, assume it is reachable and treat unauthenticated file-write as owned; the economics have changed, not the vulnerability class.

confidence 0.80

Ransomware

SafePay posted 24 victims in one day, which is a loading bay, not a crime wave

Our ransomware tempo counted 380 leak-site claims in seven days from 55 of the 281 groups tracked, with N0n surging to 92 claims against 36.19 expected from its own prior rate and Rhysida to 11 against 1.86. Underneath that, the shapes are lumpy: SafePay logged 24 items on 21 November against a median of 1 on other days, TheGentlemen 20 against a median of 0, Vortex 8, and Everest 21 on 20 November against a median of 3.5. The SafePay batch reads like a backlog being cleared, nine or ten firms across Israel, Singapore, Ecuador, Argentina, Spain, Chile, Taiwan and the Philippines posted inside fifteen minutes. For the month to date the trackers have healthcare at 116 claims, manufacturing 97 and business services 77, with the United States at 425 against Canada's 43.

Why it mattersA posting burst is a publishing decision, not an intrusion date; count distinct victims over the next two days before you tell the board that anything has escalated.

confidence 0.65

Vulnerabilities

F5 ships emergency BIG-IP patches, JetBrains ships about sixty, SonicWall ships a sentence

F5 told customers to patch five critical BIG-IP vulnerabilities that can be used for denial of service or remote code execution. JetBrains published a wall of TeamCity advisories running from CVE-2026-132891 to CVE-2026-132949, several of them remote code execution via pull requests, Subversion VCS root settings and Kubernetes Executor settings, which CERT-Bund flagged as high. And SonicWall urged everyone still running Generation 6 firewalls to retire and replace them immediately and stop exposing them at the internet edge, because they will not get security patches even for critical flaws, citing attackers who 'now use automation and AI to find and exploit unsupported edge devices'.

Why it mattersTwo of these are a patch window; the third is a procurement cycle you have not budgeted for, and the edge device is the one the attacker reaches first.

confidence 0.85

Threat actors

The FBI is investigating an apparent hack of the database the whole country runs warrants through

FBI systems have reportedly been breached for the second time in two months, this time targeting the National Crime Information Center, the database used to share information between local, state and federal law enforcement. Two people briefed on the matter told CNN the Bureau is investigating; the AP's source said the same. Reporting is thin on how, and nobody sensible is claiming to know yet.

Why it mattersNothing for you to patch, but it is the clearest live example of the thing we keep saying in slides: the shared national lookup service is a single point of trust for thousands of organisations that did not choose it.

confidence 0.75

Threat actors

Four vendor incidents in two days, and your perimeter is now somebody else's status page

N-able said attackers who breached one of its servers caused a worldwide outage of Mail Assure and SpamExperts Hosted Cloud, with possible credential theft. Rackspace warned that intruders using compromised credentials got into Rackspace Spot, deployed unauthorised workloads and accessed Kubernetes secrets in customer clusters. Microsoft told organisations that a flaw in Entra Connect Health was exploited to reach administrator email addresses and tenant configuration data. And J.P. Morgan Chase, Visa and Mastercard all pointed at a third-party provider after the N0n extortion group leaked data said to relate to their customers, while in Japan ZOZO confirmed roughly 29.13 million customer and business-partner records exposed via an attack on PLAID's KARTE platform.

Why it mattersRotate anything Rackspace Spot clusters held as a secret, review Entra Connect Health admin exposure, and accept that your third-party register is now an incident-response contact list.

confidence 0.85

AI & security

Signal turns off screenshots because the operating system is now a bystander with a camera

Signal is restricting screenshots by default on Android and iOS to stop agentic systems built into the OS from reading private chats, and has openly conceded the protection is 'incomplete'. In the same window, researchers showed that a single malicious webpage could pull a Meta Muse user's private conversations and stored memories out of the assistant with no approval prompt, and extend the same route toward credentials Muse holds for connected accounts. Meanwhile an a16z survey of CISOs at organisations with more than 1,000 employees found 36% comfortable today letting AI agents act autonomously on low-risk security actions, 60% expecting to be within twelve months, and only 7% expecting to trust AI with most actions.

Why it mattersPrompt injection against an assistant that holds memory and credentials is a credential-theft path with no malware, no login and no alert, and your DLP does not see the conversation.

confidence 0.80

Malware & IOCs

BADBOX is back at roughly 40 billion bid requests a day, and the botnets never left

The BADBOX malware found across generic Android TV boxes has returned to large-scale ad fraud after last year's disruptions, at approximately 40 billion bid requests daily and over 4% of all connected TV traffic, described as the largest ad fraud operation by bid volume ever publicly disclosed. The commodity feeds agree it is a busy week downstream: our signal collector logged 320 new Mirai indicators and 193 for Mozi in one window, 120 for SalatStealer in about half an hour, plus fresh ClickFix and Gafgyt batches, while ThreatFox absorbed a long block of Skimer card-skimming domains and a run of StrelaStealer delivery hosts on compromised legitimate sites.

Why it mattersSkimer and StrelaStealer are hosted on real businesses' domains, so blocklist-only control fails; the TV box problem is a procurement one you cannot patch.

confidence 0.80

Oddities

Everest is now extorting companies its own language model cannot confirm exist

Across 20 and 21 November, Everest's leak-site entries arrived tagged '[AI generated]', and several of them are the model declining the job: on Geesol, 'I don't have reliable, verifiable information about a company called "Geesol"'; on Accuver, 'Accuver is not a company I have reliable, verified information about'; and similar non-answers for My American Alliance, ALKF, Alla France and FAST ČR, the last noting the name is ambiguous. Elsewhere in the economy, the chief technology officer and co-founder of ransomwarehelp.com, a six-year-old Florida business that claimed to use 'advanced cryptographic tools' to restore locked data, has been charged with wire fraud after an FBI sting found the company was simply paying the attackers for a decryption key and charging a markup.

Why it mattersLeak-site victim descriptions are now partly machine-written filler, so treat the narrative as marketing and the file counts as the only claim worth checking.

confidence 0.80
§3

Hidden

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

The only story that genuinely crossed every feed was AI-agent tooling; the rest of the noise is trackers quoting each other, and three of our vulnerability sources went silent at the same minute.

  1. Exception 01

    Cairn and MONAI Label is the one real cross-feed theme

    The same story shows up in cyber-news, Telegram, Mastodon, RSS and OTX, with Telegram carrying the 'China-linked hackers' framing about 2.2 hours before the news feeds, which suggests relay channels amplify vendor reports fast. The related 'agentic' claims from Bell Canada, the European Commission and the Bazuka framework may be bundled by narrative rather than linked operationally.

    confidence 0.60
    Wrong if
    Amazon's report is retracted, or MONAI Label turns out to be patched or not unauthenticated, or the Bell Canada and EC incidents prove unrelated to autonomous frameworks.
    Watch
    A MONAI Label patch or CVE assignment, and any KEV entry, in the next few days.
  2. Exception 02

    The SafePay and TheGentlemen bursts look like batch posting

    SafePay logged 24 items on 21 November against a median of 1, TheGentlemen 20 against a median of 0, Vortex 8 against 1. Bulk posting at that shape usually reflects batched victim uploads or scraper re-indexing rather than a real surge in intrusions. TheGentlemen is the weakest corroborated of the three, visible only in atlas and its derived signal, with atlas first by about 18 hours.

    confidence 0.50Speculative
    Wrong if
    The posted victims carry earlier dates or duplicate across items, which would show a re-index; or the burst persists as many distinct new victims on later days.
    Watch
    Distinct victim counts for these three groups on 22 and 23 November.
  3. Exception 03

    NVD, CISA-KEV and OTX all stop at 2026-11-20T14:44

    Three unrelated sources going quiet at the same minute is a collection or ingest problem, not a quiet weekend. Absence of CVE or KEV news after that timestamp carries no information, and the later days in this window are biased toward ransomware trackers and social feeds.

    confidence 0.65Speculative
    Wrong if
    The next collection run returns NVD, KEV or OTX items with event timestamps after 2026-11-20T14:44.
    Watch
    The next ingest for those three sources.
  4. Exception 04

    KEV entries reached the news desks in about ten hours

    CVE-2026-123890 and CVE-2026-76466 appeared in KEV roughly 9.6 and 10.5 hours before cyber-news coverage, and CVE-2026-129448 about 28.4 hours before Mastodon picked it up. Three CVEs is a description of this window, not a rate.

    confidence 0.40
    Wrong if
    The item timestamps are scrape times rather than event times, which would invalidate the lags.
    Watch
    How fast the next KEV additions are covered.

Nothing to seeNo timing clusters were detected at all. Most of the apparent corroboration between ransomlook, ransomware.live and signal:atlas is the same post travelling between mirrors, so counting it as independent confirmation would be flattering ourselves; and the Telegram-first lead-lag pairs rest on two items each, which is a coincidence with a stopwatch.

Deep read: 811 items read

§4

Forecast

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

Today’s calls

  1. DueSun 29 Nov7d

    SafePay posts fewer than 10 new leak-site claims in the seven days to 29 November, consistent with 21 November being a batch upload rather than a surge in intrusions.

    Wrong ifSafePay logs 10 or more distinct new victims across the trackers in that window.

    f-2026-11-22-1

    Confidence0.60
  2. DueSun 29 Nov7d

    N0n keeps running hot: at least 40 further leak-site claims in the seven days to 29 November, having already posted 92 against an expected 36.19.

    Wrong ifN0n posts fewer than 40 claims in that window.

    f-2026-11-22-2

    Confidence0.55
  3. DueWed 2 Dec10d

    No MONAI Label vulnerability appears in the CISA KEV catalogue by 2 December, despite root-level exploitation being publicly reported.

    Wrong ifCISA adds a MONAI Label CVE to KEV on or before 2 December.

    f-2026-11-22-3

    Confidence0.80

Scorecard

Noted

Calls made today are graded here from tomorrow. Including the wrong ones. Especially the wrong ones.

Glossary of terms you will be expected to nod at

Agentic (n.)
Of an intrusion: performed by software that did not get bored, applied to a vulnerability that a bored human would also have found, and described in the statement because it sounds better than 'it ran overnight'.
§5

From our edge

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

Our own edge, last 24 hours: the sensor profiled 102 hostile sources — 58 scanners, 42 bot crawlers, two brute forcers — which between them sent 11,700 requests, of which 52 went past looking and into exploitation attempts, across 6,371 addresses from 120 countries, 2,392 of them seen for the first time today. Eight groups of addresses shared a behavioural fingerprint, the largest being nineteen IPs all running the same tool under DNA ce6ecdfacbe3, asking for /.git/config, /config/gcp.json and an AWS credentials file by path traversal. One scanner, 161.120.215.162, fired 167 requests in about eight seconds hunting .env files, which is not stealth, it is a vending machine. Two addresses tried CVE-2022-26134 at /shell, a named Confluence bug from a year people have stopped mentioning, and the top hostility score of the day, 6.8, went to a host sending 135 requests for cache.php nested inside ever more /images/ directories, as if the file might be hiding further in. Raw logs never left the server; only these findings did.

102hostile sources profiled
11,700requests from them
52reached exploitation
8shared-DNA clusters
8.2/10top hostility
6,371addresses at our edge
2,392first seen today
120countries

The WTF list

  1. WTF #1

    The most determined visitor

    77.67.9.210, a scanner (hostility 6.8/10), sent 135 requests to one of our sites, probing for /images/images/images/images/cache.php, /images/images/images/cache.php, /images/images/images/images/images/cache.php, /assets/images/images/cache.php.

    Why it’s weirdHighest hostility on the board; its kill chain reached weaponization.

    attribution: unknownSat 21 Nov, 17:00 UTC

    Detail
    kill chain: reconnaissance → weaponization · DNA dfef4501b216
  2. WTF #2

    Same DNA, different addresses

    91.92.47.77, a scanner (hostility 4.9/10), sent 525 requests to 2 of our sites, probing for /.git/config, /config/vertex.json, /config/gcp.json, /../../../../../../home/ubuntu/.aws/credentials.

    Why it’s weird19 IPs share one behavioural fingerprint (DNA ce6ecdfacbe3): one tool or one operator rotating addresses: 34.14.60.178, 207.175.42.194, 34.77.15.40, 35.187.179.124, 34.178.6.65 and more.

    attribution: unknownSat 21 Nov, 17:00 UTC

    Detail
    kill chain: reconnaissance → weaponization → exploitation · DNA ce6ecdfacbe3
  3. WTF #3

    Trying CVE-2022-26134 on us

    2 addresses sent 2 requests shaped like CVE-2022-26134 exploitation (/shell), including 160.187.22.20, 5.78.119.228.

    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: confluence-ognl
  4. WTF #4

    The strangest shopping list

    4.197.158.170, a scanner (hostility 6.6/10), sent 20 requests to one of our sites, probing for /wp-avac.php, /wp-content/plugins/hellopress/wp_filemanager.php, /this_is_a_new_hello_world.php, /aw.php.

    Why it’s weirdAsked for files almost nobody else asks for: /wp-avac.php, /wp-content/plugins/hellopress/wp_filemanager.php, /this_is_a_new_hello_world.php, /aw.php, /images.php.

    attribution: WordPress Plugin Vulnerability Scanner (HelloPress CVE) (sensor's best match, 48%)Sun 22 Nov, 05:00 UTC

    Detail
    kill chain: reconnaissance → weaponization · DNA dfef4501b216
  5. WTF #5

    In a hurry

    161.120.215.162, a scanner (hostility 6.6/10), sent 167 requests to one of our sites, probing for /.git/config, /.env, /.env.local, /.env.production.

    Why it’s weirdFired 167 requests in about 8.0 seconds.

    attribution: Automated .env credential-harvesting scanner (sensor's best match, 65%)Sun 22 Nov, 05:00 UTC

    Detail
    kill chain: reconnaissance → weaponization · DNA dfef4501b216
  6. WTF #6

    Somebody we might know

    157.230.254.35, a scanner (hostility 5.9/10), sent 24 requests at 2 a minute to 2 of our sites, probing for //images/images/cache.php, //cgi-bin/cgi-bin/cgi-bin/cgi-bin/cache.php, //images/images/images/images/images/images/images/images/cache.php, //images/images/images/images/images/images/images/cache.php.

    Why it’s weirdThe sensor's most confident campaign match of the day: PHP cache.php vulnerability mass-scan campaign at 52%.

    attribution: PHP cache.php vulnerability mass-scan campaign (sensor's best match, 52%)Sat 21 Nov, 17:00 UTC

    Detail
    kill chain: reconnaissance → weaponization · DNA bccc30d19899
§6

Tempo

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

380claims, 7 days
55of 281 groups active
N0n92 this week · surging
SoulGroup27 this week · new
Vortex26 this week · new
  • SurgingN0n — 92 claims in 7 days against 36.2 expected from its own prior rate (p < 0.001)
  • Surgingrhysida — 11 claims in 7 days against 1.9 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)
  • Surgingeverest — 10 claims in 7 days against 3.0 expected from its own prior rate (p = 0.001)
  • Surgingaurora — 7 claims in 7 days against 1.9 expected from its own prior rate (p = 0.004)
  • Surgingpear — 6 claims in 7 days against 1.9 expected from its own prior rate (p = 0.015)

Gone quiet: titan (61 days), moneymessage (62 days), unsafe (62 days), killsec (65 days).

Sector pressure, 30 days

  • UpHealthcare — 1.3× its own prior share (162 claims, 123 expected)
  • UpRetail & E-Commerce — 1.3× its own prior share (107 claims, 84 expected)
  • DownAgriculture and Food Production — 0.3× its own prior share (15 claims, 46 expected)
  • DownTransportation — 0.7× its own prior share (38 claims, 57 expected)

25% of claims carry no sector label.

Brand churn

Of the 243 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 218 old enough to judge posted on one day and never again; 38% made at least half their claims in their first week. New brands this month: 12 — newest: SoulGroup (2026-11-19, 27 claims), SilentVault (2026-11-19, 8 claims), Handsome Boy Hacking Club (2026-11-19, 4 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: AuditTeam put 32% of its claims on Russia (169.1× every other brand's share); safepay put 31% of its claims on Germany (6.0× 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 10.6 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: clop released 94% of its claims on Wednesdays, half of them in 13:00–15:59 UTC; Deadlock released 60% of its claims on Fridays, half of them in 12:00–13:59 UTC; akira posted only 2 of 196 claims at weekends and half its claims in 12:00–13:59 UTC; SilentRansomGroup posted only 2 of 92 claims at weekends and half its claims in 19:00–22:59 UTC. Leak sites overall are 39% 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 56 was made by machine and checked by code. Every item published by 14 public threat feeds in the 24 hours to 05:05 UK (811 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 (0 today). Forecasts are graded in public in later editions. 'From our edge' is live attack data from our own servers. Cost of this edition: $1.41. 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.

811items read
24hour window
$1.41cost of this edition
05:05 UTCgenerated

Sources this morning

  • NVD200ok
  • Telegram127ok
  • ThreatFox100ok
  • Signal Atlas91ok
  • infosec.exchange75ok
  • ransomware.live56ok
  • RansomLook54ok
  • Security news48ok
  • atlas26ok
  • RSS12ok
  • AlienVault OTX12ok
  • signal:ioc-family5ok
  • CISA KEV4ok
  • Ransomware tempo · Ninja Signal graph1ok

ITEMS READ: 811SOURCES: 14STORIES DROPPED FOR LACK OF RECEIPTS: 0COST: $1.41STATUS: NOT ESPECIALLY FINE

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