Case file — Lead incident: Apify, the Czech web-scraping and automation platform, has confirmed that an unauthorized third party accessed parts of its internal systems on November 9–10, 2026, and reached customer personal information and stored data. The company detected the intrusion on November 11 at 12:13 UTC, cut off the attacker's access, and began rotating customer API tokens and proxy passwords, resetting sessions, revoking GitHub OAuth authorizations, and asking customers to reset two-factor authentication. Apify has not said how many customers were affected — the platform lists about 80,000 customers — and has not named an attacker or described how the intruder got in. On November 13 the company said API tokens are now stored only as hashes, that it has added network restrictions between internal services and stronger internal authentication, and that a root-cause post-mortem will follow within a week. A breach notification shared by a customer and cited by SecurityWeek lists names, email addresses, billing details, account activity, IP addresses, salted password hashes, and 2FA recovery codes among the exposed data — Apify itself has not confirmed that list. Status: containment and credential rotation are underway; the incident was still listed as unresolved at the time of writing, and customers should treat exposed tokens and passwords as compromised.
Incident timeline
| Date | Development |
|---|---|
| November 9–10, 2026 | An unauthorized third party gains access to parts of Apify's internal systems. Separately, the Check Point campaign disclosed this week is already underway — its first wave began October 30 against Israeli organizations, with a second wave on November 2. |
| November 10, 2026 | Security researcher Abdelhamid Naceri publishes the SignDrop Windows zero-day proof of concept on GitHub, the same day as Microsoft's November Patch Tuesday. Microsoft's Exchange Online infrastructure begins failing in the Americas in the evening — the start of the Microsoft 365 outage. |
| November 11, 2026, 12:13 UTC | Apify detects the intrusion, cuts off access, and begins rotating credentials; its first public status notice follows at 20:46 CET. Trezor releases firmware 2.13.1, which blocks the hardware implants found in three Safe 5 wallets from capturing new data. |
| November 12, 2026 | Check Point discloses the PHP2Shell-to-ESXi-wiper campaign against its Security Gateways. Trezor publicly confirms the three implanted Safe 5 wallets. BleepingComputer reports the SignDrop zero-day. The Microsoft 365 outage spreads worldwide through the working day in Asia and Europe. |
| November 13, 2026 | CISA adds Check Point's CVE-2026-123039 to its Known Exploited Vulnerabilities catalog, with a federal remediation deadline of Monday, November 16. Apify says tokens are now stored only as hashes and promises a post-mortem within a week. Microsoft apologizes for the outage and says services are broadly recovering; Downdetector logged a peak of about 138,000 Outlook problem reports. |
| November 14, 2026 | Microsoft says affected Microsoft 365 services have largely recovered, with a backlog of delayed email still draining and no ETA yet for full recovery. The company maintains the outage was a technical failure, not a cyberattack. |
1. Apify breach exposed customer data; tokens rotated, 2FA resets urged
Apify, a Czech web-scraping and automation cloud platform with about 80,000 customers, has confirmed a breach of its internal systems. An unauthorized third party gained access to parts of those systems on November 9–10, 2026, and Apify later confirmed that customer personal information or stored data was accessed. The company detected the activity on November 11 at 12:13 UTC, cut off the attacker's access, and began an investigation the same day.
The response has been a full credential reset. Apify rotated customer API tokens and Apify Proxy passwords, reset active sessions, revoked GitHub OAuth authorizations, and told customers to reset their two-factor authentication. On November 13 the company said API tokens are now stored only as hashes, that it has added network restrictions between internal services and stronger internal authentication, and that it has begun a review of its whole stack. A root-cause post-mortem was promised within a week, and the incident was still listed as unresolved at the time of writing.
Exactly what data was taken remains partly unclear. SecurityWeek, citing a breach notification a customer shared publicly, lists names, email addresses, billing details, account activity, IP addresses, salted password hashes, and 2FA recovery codes — but Apify has not confirmed that list itself, so treat it as reported rather than established. Apify has not said how many customers were affected, has not named an attacker, and has published no attacker indicators, which makes detection behaviour-first: any unexpected client reaching the Apify API or proxy should be reviewed, and any token or password that existed before November 11 should be treated as exposed.
Investigation notes
What is confirmed: unauthorized access on November 9–10, detection on November 11 at 12:13 UTC, customer personal information or stored data accessed (Apify, via its status page and BleepingComputer, November 12–13); the token and proxy-password rotation, session resets, GitHub OAuth revocation, and 2FA reset guidance; the November 13 hardening steps and the promised post-mortem.
What is not confirmed: the detailed data list — names, billing details, salted password hashes, 2FA recovery codes — comes from a customer-shared notification cited by SecurityWeek, not from Apify. Per the evidence rule, report it as reported, not as established fact. The attacker, the entry point, and the number of affected customers are all still undisclosed.
What customers should do: rotate every Apify API token and proxy password, reset 2FA, revoke and re-authorize OAuth integrations, and review API and proxy usage for sessions or clients that do not belong. The customer names on Apify's marketing page — Microsoft, Samsung, T-Mobile, Accenture, Siemens, the European Commission, Roche — are not stated to be affected.
2. Microsoft 365 outage knocked out Outlook, Teams, and Copilot for nearly a day
Microsoft's cloud productivity services went dark worldwide for nearly a day in one of the largest Microsoft 365 outages in years. The disruption began on the evening of Thursday, November 12, in the Americas and ran through the working day in Asia and Europe on Friday, November 13, hitting Outlook email, Microsoft Teams, Copilot, SharePoint, OneDrive, Planner, and To Do across at least 70 countries. Downdetector logged a peak of about 138,000 Outlook problem reports around 1 p.m. ET on Thursday. Microsoft apologized on Friday and said services were broadly recovering; by early Saturday it said it had addressed the underlying issue but was still draining a backlog of delayed email, with no ETA for full recovery and the incident not yet declared resolved.
Microsoft says this was not a cyberattack. The company traced the failure to an issue during an earlier network capacity upgrade that surfaced after a recent network change: a process responsible for establishing secure connections became overwhelmed on affected Exchange Online servers, which then stopped processing requests. Microsoft was explicit that the incident was "not related to cyber or AI/agents", and said it found no evidence of malicious activity. A full post-incident report has been promised once its internal review concludes.
The outage is a resilience story rather than a security breach, but it belongs in this brief for the same reason defenders track it: for most of a working day, the email, calendar, and messaging stack that incident response itself runs on was unavailable. Organizations that run their security operations through Outlook and Teams should treat the episode as a prompt to check what their incident communications look like when those tools are the thing that is down.
Investigation notes
What is confirmed: the outage window (evening of November 12 through November 13, nearly a day), the affected services, the 70-plus-country footprint, the Downdetector peak of roughly 138,000 Outlook reports, Microsoft's stated cause, and the explicit ruling-out of a cyberattack. Microsoft 365 was the affected product — Azure was not listed in the incident, despite a spike in Azure complaints on tracking sites.
What is still thin: exact user numbers and the full chain of events are not public yet — those wait on Microsoft's post-incident report, for which no date has been given. Published durations vary between "nearly 18 hours" and "more than 24 hours" depending on the outlet and the recovery milestone it counts from; "nearly a day" is the honest summary.
What is not in doubt: this was a technical failure, not an attack. Any claim tying the outage to a threat actor or to AI agents contradicts Microsoft's own statement and the evidence published so far.
3. Check Point gateways hacked via PHP2Shell to deploy VMware ESXi wiper
Check Point has disclosed that attackers chained the PHP2Shell flaws with a newly discovered zero-day to compromise its Security Gateways, plant a backdoor, and — in some cases — wipe VMware ESXi hosts. The campaign began on October 30, 2026, against Israeli organizations, and a second wave on November 2 reached companies in multiple countries and sectors. Check Point says it observed active exploitation against a couple of dozen customers.
The attack chain starts at the gateways' internet-facing Mobile Access portal, where the attackers exploited the PHP2Shell vulnerabilities (CVE-2026-109618 and CVE-2026-109619) — the critical PHP remote code execution flaws patched in emergency releases on October 21 and added to CISA's Known Exploited Vulnerabilities catalog on October 26. From there they used CVE-2026-123039, a previously unreported local privilege escalation in the Gaia OS DynamicCLI daemon (CVSS 7.8), to gain root. The daemon runs as root, listens on a local socket any process on the box can write to, and never checks who is connecting. The attackers then installed PHP webshells and the TSGA backdoor, which embeds a networking component called TSProxy that hides attacker commands inside regular traffic.
In the first wave, the attackers pivoted from the gateway to VMware ESXi hosts found with the open-source gogo scanner, and deployed a data wiper called destroy_v6 that recursively deletes data in the target directory, leaving a note named DREAM_ANNEXATION.txt that references "Dream Security Ltd." Check Point tentatively linked TSGA and TSProxy to PhiliKit, a malware framework previously seen in attacks on Ivanti appliances and associated by ESET with the Chinese threat cluster UNC5221 — but the company says it does not have enough evidence to attribute the campaign with confidence. CISA added CVE-2026-123039 to its KEV catalog on November 13, with a federal remediation deadline of Monday, November 16. The fix is the November Jumbo Hotfix on all Management Servers and Security Gateways, prioritizing internet-facing Mobile Access portals.
Investigation notes
Official anchor: Check Point's November 12 advisory and CISA's KEV listing of CVE-2026-123039 on November 13, due November 16. The PHP2Shell flaws were already KEV-listed on October 26. This is the same PHP2Shell family this brief covered on October 23 — now with a confirmed victim chain, a second zero-day, and a destructive final stage.
Attribution, stated carefully: the PhiliKit and UNC5221 overlap is a tentative technical link, not an attribution. Check Point explicitly says it cannot attribute the campaign with confidence, and the DREAM_ANNEXATION.txt note is a data point, not a signature. Do not present this as a China-attributed campaign.
What is confirmed: the two-wave timeline, the entry and escalation chain, the TSGA/TSProxy backdoor, the ESXi pivot and destroy_v6 wiper in first-wave victims, and the "couple of dozen" observed customer count — a floor, not a total.
What gateway operators should do: install the November Jumbo Hotfix (R82.20 Take 14, R82.10 Take 52, R82 Take 132, or R81.20 Take 174 or later) on every Security Gateway and Management Server, internet-facing Mobile Access portals first. If a gateway looks compromised, treat any ESXi host it could reach as suspect and confirm offline backups exist before cleaning up.
4. Trezor confirms crypto-stealing implants in three wallets, releases fix
Hardware wallet maker Trezor has confirmed that three of its Safe 5 devices reached customers with hidden spy implants inside them — extra circuit boards carrying cellular modules and antennas, fitted between the main board and the screen, designed to copy the wallet's recovery phrase and any displayed passphrase, encrypt them, and send them to an attacker over a mobile network. All three were bought from CryptoBilis, a Southeast Asian reseller, between August 20 and September 23, 2026, through the reseller's own store and its shops on Tokopedia and Shopee.
Trezor examined one device in its laboratory and identified the other two from photographs submitted by their owners; 48 devices from CryptoBilis customers were inspected in total, and only the three were confirmed tampered. The company released firmware 2.13.1 on November 11, a day before the public disclosure, which changes the order of the commands sent to the display so the implant's trigger sequence no longer appears — Trezor says this makes the current batch of implants useless. But the fix cannot protect a recovery phrase that was already captured before it was installed, so CryptoBilis buyers are still advised to move funds to a newly created wallet with a fresh recovery phrase. Trezor has cut its authorized reseller network by about 80 percent, is offering replacement devices or refunds to anyone who bought from CryptoBilis since March 2026, and is working with crypto incident-response group SEAL 911. No customer has reported losing funds, and no implants have been found in devices from other resellers.
The investigation began with a responsible disclosure on October 28 flagging possible tampering in a Safe 5 bought in August, and Trezor ended its relationship with CryptoBilis on October 9. The same reseller has also been linked in reporting to tampered Ledger devices, in a case where an estimated $72 million to $102 million was drained — that figure is an estimate from other reporting, not a Trezor finding.
Investigation notes
Why this matters beyond crypto: this is a physical supply-chain compromise — the tampering happened between the factory and the customer, at the reseller. It is the hardware equivalent of a poisoned software update, and the defence is the same shape: buy from the vendor or a verified channel, verify the device, and treat a third-party seller as untrusted until proven otherwise.
What is confirmed: three implanted Safe 5 devices, all from CryptoBilis, bought August 20–September 23, 2026 (Trezor, November 12); the implant design and its cellular exfiltration path; firmware 2.13.1 and what it does and does not protect; the ~80 percent reseller cut; no reported fund losses.
What is not known: who built and fitted the implants, how many devices from the reseller were tampered with beyond the three confirmed, and whether any recovery phrase was captured before the firmware fix shipped. "No customer has reported losing funds" is Trezor's statement — an implant that stole a phrase would leave no trace until the thief moved the funds.
5. New Windows 'SignDrop' zero-day gives attackers SYSTEM privileges
Security researcher Abdelhamid Naceri, known as Nightmare Eclipse, has published a working proof-of-concept exploit for a new, unpatched Windows privilege-escalation flaw that hands a local attacker full SYSTEM rights. Named SignDrop, it was released on November 10 — the same day as Microsoft's November Patch Tuesday, which fixed 1,093 flaws — and targets the Common Log File System (CLFS) driver.
SignDrop abuses the very mitigation Microsoft introduced in October 2025 to defend CLFS against authentication attacks. A standard user prepares a CLFS log and container, then uses Object Manager symbolic links to redirect where the driver's .cnpf authentication file lands. When a privileged process such as Windows Defender later opens the prepared log — coerced through its signature-update RPC interface — CLFS creates the authentication file in that process's SYSTEM context, at the attacker-chosen path, with the original file's access permissions. The result is a SYSTEM-owned file the standard user can write to, and a SYSTEM shell. Naceri tested the exploit on Windows 11 26H2, an experimental 26H2 build, and an experimental Windows Server 2025 build, and estimates that about 80 percent of Windows installations are vulnerable — that figure is the researcher's estimate, not a confirmed count, and independent reproduction by researcher Will Dormann confirmed the exploit works as described, though not on all Windows 11 versions.
There is no CVE, no patch, and no reported in-the-wild exploitation, and the flaw is not in CISA's KEV catalog. Microsoft has not commented on SignDrop. Naceri notes there is currently no solution except waiting for a patch: disabling the CLFS mitigation would re-expose machines to the memory-corruption attacks it was built to stop. Machines can be checked by running the proof of concept, or by reading the CLFS authentication mode in the registry — a value of 0 or 1 (learning or enforcement mode) means the machine is in scope.
Investigation notes
Evidence rule, applied: SignDrop is an unpatched zero-day with a public proof of concept, not an exploited vulnerability. No in-the-wild abuse has been reported and no CVE exists. Report the capability as demonstrated, the exposure estimate as the researcher's own, and the exploitation status as unproven.
The irony, stated plainly: the mitigation Microsoft shipped in October 2025 to stop CLFS authentication attacks is itself the attack surface. Naceri says attacks exploiting CLFS vulnerabilities dropped sharply after the mitigation — and that the mitigation introduced this new flaw. It is a reminder that a defensive change is also a code change, and it needs the same scrutiny as the bug it replaces.
Scope, with the caveat attached: the ~80 percent figure is Naceri's estimate. Dormann's reproduction adds an important qualifier: the vulnerable machines appear to be Windows 26H2 installations that have existed for more than 90 days, which narrows the real population considerably given Microsoft's staged rollouts. Do not quote 80 percent without that qualifier.
Sources
- Apify rotates API tokens, prompts 2FA resets after breach — BleepingComputer
- Microsoft apologises after global outage disrupts Outlook, Teams and Copilot for nearly 18 hours — The Hindu BusinessLine
- Check Point gateways hacked via PHP2Shell to deploy VMware ESXi wiper — BleepingComputer
- Trezor confirms crypto-stealing implants in three wallets, releases fix — BleepingComputer
- New Windows 'SignDrop' zero-day gives hackers SYSTEM privileges — BleepingComputer
- Check Point patches zero-day that lets hackers get root access — BleepingComputer
