Monday, 7 September 2026

Dissecting a PHP web server rootkit (opens in a new tab)

Sophos Threat Research ·fetched 7 Sep 2026, 11:41 UTC Must read Research agreed3/3

Why readReverse-engineers a PHP rootkit that hooks Apache via libphp and APR module loading to survive F5 BIG-IP upgrades, tied to exploitation of CVE-2025-53521 in BIG-IP APM.

The implant gives on-demand server-side code execution like a web shell but does it through Apache and libphp module loading rather than a dropped script, and is purpose-built for BIG-IP APM webtop and upgrade workflows. Analysis puts the sample as a second stage; a related umount sample revealed an installer that infects /usr/sbin/httpd, persists across BIG-IP upgrade images and modifies SELinux configuration. F5 links the c05d5254 activity to BIG-IP APM systems hit by CVE-2025-53521, an exploited unauthenticated RCE where an access policy is bound to a virtual server, and Sophos tells affected operators to follow F5's compromise-assessment guidance before generic Apache or PHP hardening.

sophos.com → (opens in a new tab)

Shai-Hulud Rises From the Dead after 111 days (opens in a new tab)

Aikido Security ·fetched 7 Sep 2026, 15:40 UTC Must read Research agreed3/3

Why readA byte identical npm payload that the whole industry fingerprinted in May was accepted by the registry again 111 days later, which says publish time malware scanning is not catching known hashes.

Aikido's triage queue surfaced a package carrying SHA-256 e37e3ddeeaaa9e0c4fdbcb829b4895a6521031c80053fc436625b61e6ee5b1a6, the same file used in the 19 May 2026 Shai-Hulud wave that pushed 639 malicious @antv versions to npm in an hour from a compromised maintainer account. The payload is unchanged, not a variant, and it went through despite npm introducing publish time malware scanning after the original incident. The 111 day gap between takedown and republication is the operational point: dormancy defeats detection windows tuned to recent activity, so pin the hash in your own registry proxy and blocklist rather than relying on the upstream scanner.

aikido.dev → (opens in a new tab)

Online Maths Learning Platform Mathspace Disclosed Data Breach Impacts 1 Million Users (opens in a new tab)

Cybersecurity News ·Guru Baran ·fetched 7 Sep 2026, 15:40 UTC Must read CVE-2026-72898 EPSS 94.2% agreed3/3

Why readProof that the unauthenticated SQL injection in self-hosted Metabase is being used in anger, with a million record breach as the result.

Mathspace disclosed on 3 September 2026 that attackers reached an internal reporting system and downloaded records on 1,079,819 students, parents, guardians, teachers and its own staff across Australia and New Zealand. The way in was CVE-2026-72898, an unauthenticated SQL injection in the password-reset API endpoint of the company's self-hosted Metabase business intelligence install, a bug now sitting at an EPSS of 0.94. Treat any internet-reachable Metabase instance as a live target rather than an internal reporting nicety.

Also covered byBleepingComputer (opens in a new tab),ABC News & Headlines – Australian Broadcasting Corporation (opens in a new tab),Cyber Daily (opens in a new tab),Qazinform (opens in a new tab),Nine.com.au (opens in a new tab),tech-insider.org (opens in a new tab).

cybersecuritynews.com → (opens in a new tab)

New Ted Backdoor Hides Inside Victims' Own HAProxy Builds to Intercept Web Traffic (opens in a new tab)

The Hacker News ·The Hacker News ·fetched 7 Sep 2026, 07:38 UTC Must read agreed3/3

Why readRapid7 documents 'ted', a Linux implant compiled into victims' own HAProxy builds that hides its C2 by decrementing HAProxy's live connection counters so neither backend logs nor load balancer stats record the traffic.

Two South Korean organisations, in the automotive and media sectors, were found running trojanized HAProxy load balancers with an undocumented toolkit named ted in its debug strings. The implant triggers C2 mode on a request for one specific image path, never forwards those requests to a backend, and decrements HAProxy's own connection counters so the session disappears from the balancer's statistics. Rapid7 attributes it to North Korean state actors with medium confidence; it is not an HAProxy vulnerability and requires prior code execution plus binary replacement, so integrity monitoring of load balancer binaries is the control that catches it.

thehackernews.com → (opens in a new tab)

North Korean Hackers Deploy New Linux Espionage Toolkit (opens in a new tab)

SecurityWeek ·Ionut Arghire ·fetched 7 Sep 2026, 19:38 UTC agreed3/3

Why readA North Korea-linked Linux implant compiled into HAProxy 2.8.12 itself, using its filter API and memory pools to intercept traffic, along with the trojanised binaries to check for.

Rapid7 documents a Linux espionage framework used against automotive and media targets in South Korea, built around a 'ted backdoor' compiled as part of the victim's own HAProxy 2.8.12 build plus trojanised agetty, atd, crond, polkitd and sshd. The backdoor rides HAProxy's native filter API, memory pools, event scheduler and process management to intercept traffic and evade monitoring while legitimate load balancing continues normally, and supports remote command execution, credential harvesting and script injection into web traffic. Likely in use since late 2024, which gives hunters a window to look back over on internet-facing load balancers.

securityweek.com → (opens in a new tab)

BigBear Microsoft 365 phishing service bypassed MFA at 258 organizations (opens in a new tab)

BleepingComputer ·Bill Toulas ·fetched 7 Sep 2026, 23:41 UTC agreed3/3

Why readAn Evilginx2-based phishing-as-a-service kit stole over 5,000 Microsoft 365 sessions across 258 organisations by proxying MFA, so cookie replay is the detection target, not password reuse.

CloudSEK obtained administrator access to the BigBear 2.0 control panel and found 42 VPS nodes all configured against Microsoft 365. The kit's "offy" configuration runs an adversary-in-the-middle proxy between victim and Microsoft's real authentication endpoints, capturing credentials plus post-MFA session cookies and replaying them through an API to hijack the session. Push and OTP factors do not stop this; phishing-resistant factors, token binding and session-anomaly detection do.

bleepingcomputer.com → (opens in a new tab)

PEEP Turns Chrome and Edge Into Post-Compromise Backdoors for Host Command Execution (opens in a new tab)

The Hacker News ·The Hacker News ·fetched 7 Sep 2026, 23:41 UTC agreed3/3

Why readDetails how PEEP forges Chromium Secure Preferences integrity values to sideload an unlisted extension, then bridges to host command execution via native messaging, with C2 IPs and a 30 second polling interval to hunt on.

SOCRadar's analysis, relayed here, describes a post-compromise toolkit that injects an extension directly into Chrome and Edge profiles by forging Chromium's own Secure Preferences HMAC values, bypassing Web Store checks and user prompts; it requires prior admin or code execution. The agent beacons to 206.237.30[.]232 and xfjcc[.]fun over plaintext HTTP every 30 seconds, exfiltrating history, active-tab metadata and session cookies, while a native-messaging host extends it to host commands and file management. PEEP builds on the open-source RedExt red-team framework, previously seen in GlassWorm activity, adding installer routines, heartbeat telemetry and an update channel. Detection angles: Secure Preferences integrity mismatches, unexpected native messaging host registrations, and regular 30 second plaintext beacons.

thehackernews.com → (opens in a new tab)

How a hole in Lenovo’s login system let hackers walk into 5,000 Dropbox accounts (opens in a new tab)

Graham Cluley ·Graham Cluley ·fetched 7 Sep 2026, 11:41 UTC agreed3/3

Why readA concrete, cheap account takeover path built entirely out of one identity provider's missing email verification plus a forgotten federation trust.

Dropbox has told roughly 5,000 users that their accounts were accessed between 4 and 21 August after attackers registered fresh Lenovo IDs against victims' email addresses. Lenovo's registration flow did not confirm the registrant controlled the inbox, and a legacy Lenovo ID integration on the Dropbox side accepted that identity as sufficient to log in with no Dropbox password. The lesson worth carrying into your own estate is that every dormant federation trust inherits the weakest verification step of the partner IdP, so inventory the ones nobody has looked at since they were built.

bitdefender.com → (opens in a new tab)

JSCeal Malware Can Bypass Google Authentication Using Stolen Session Cookies (opens in a new tab)

The Hacker News ·The Hacker News ·fetched 7 Sep 2026, 11:41 UTC agreed3/3

Why readBreaks down JSCeal's compiled V8 (JSC) payloads, the javascript-obfuscator protections used (RC4 strings, control-flow flattening, proxy functions), and its theft of Google session cookies to sidestep authentication.

Check Point's technical report unpacks JSCeal, a compiled-V8 JavaScript implant delivered through fake TradingView installers on counterfeit crypto trading sites promoted via Facebook and Google ads. Delivery uses two PowerShell-fetched ZIP archives, one carrying a portable Node.js runtime and one the payload, and the malware performs credential harvesting, surveillance and traffic interception, including reuse of stolen session cookies against Google accounts. Activity overlaps clusters tracked as WEEVILPROXY and MeadowLocust, first documented by Check Point in July 2025.

thehackernews.com → (opens in a new tab)

Detection and Removal of the Syslogk Rootkit in a Linux Environment (opens in a new tab)

AhnLab ASEC ·ATCP ·fetched 7 Sep 2026, 03:42 UTC agreed3/3

Why readWalks the two hooking mechanisms Syslogk uses to hide processes, TCP connections and files, naming the kernel API functions patched by each.

Analysis of the Syslogk Linux kernel rootkit, which combines inline hooking (patching the prologue opcodes of target kernel functions to redirect execution) with VFS table hooking to conceal processes, network communications and files on disk. The write-up maps which API functions each technique targets and what each hook is hiding, then moves to detection and removal. It is a competent explanation of a known rootkit rather than new discovery, and the remediation half leans on AhnLab's own products, but the hooking breakdown is usable for anyone writing Linux kernel integrity checks.

asec.ahnlab.com → (opens in a new tab)

North Korea’s Lazarus Operates Through Six Distinct Cyber Clusters (opens in a new tab)

Infosecurity Magazine ·fetched 7 Sep 2026, 11:41 UTC agreed3/3

Why readBreaks the Lazarus umbrella into six named clusters, TEMP.Hermit, Citrine Sleet, CryptoCore, Jade Sleet, Moonstone Sleet and Famous Chollima, mapped mostly under North Korea's GRIB.

Sekoia and Kudelski Security published a TTP-based clustering on 7 September that splits the former Lazarus label into six distinct groups, separated by mission: espionage, revenue generation and sanctions evasion. Most of the units sit under GRIB, the military intelligence bureau formerly known as the RGB, and the researchers note that repeated internal reorganisation and renaming is what makes attribution to a single Lazarus entity misleading. Useful for CTI teams reconciling vendor naming, though this is the news write-up rather than the underlying report.

infosecurity-magazine.com → (opens in a new tab)

Phishing Campaign Sends Millions of Emails Using Invisible Unicode to Evade Filters (opens in a new tab)

The Hacker News ·The Hacker News ·fetched 7 Sep 2026, 03:42 UTC agreed3/3

Why readShows attackers splitting financial lure words such as 'funding' with invisible Unicode tag characters so email filters fail to parse them, a repurposing of ASCII smuggling for ordinary spam evasion.

Microsoft Security Research reports a high-volume phishing campaign, first seen in early February 2026, that inserts non-rendering Unicode tag characters inside lure words to defeat keyword-based email filtering while the text renders normally to the recipient. The same ASCII smuggling primitive has been used against LLMs for prompt injection; here it is aimed at the filter rather than the model. Detection work follows directly: normalise or strip Unicode tag-block codepoints before content inspection.

thehackernews.com → (opens in a new tab)

Magento StyleSmuggler zero-day exploited to deploy Linux backdoor (opens in a new tab)

BleepingComputer ·Bill Toulas ·fetched 7 Sep 2026, 19:38 UTC Must read agreed3/3

Why readUnauthenticated RCE in all current Magento and Adobe Commerce versions, exploited in the wild since 4 September with no patch, plus the process names and cron interval to hunt for.

Sansec reports a zero-day dubbed StyleSmuggler being exploited against Magento and Adobe Commerce, with the first case on 4 September against a fully patched store. The exploit injects PHP through Magento's template system to generate a fake failed-payment email that triggers code execution, then drops a small Rust backdoor masquerading as [kworker/u:8:0], and in newer builds as fc-cache copied to ~/.cache/fontconfig/fc-cache, with a cron job every 30 minutes for persistence. Adobe has confirmed it is working on a fix but has given no timeline; roughly 160,000 sites run the platform.

bleepingcomputer.com → (opens in a new tab)

Back-to-back N-able bugs send admins on a patching spree (opens in a new tab)

CSO Online ·fetched 7 Sep 2026, 15:40 UTC CVE-2026-86207 EPSS 0.4% agreed3/3

Why readThe clearest account of the N-central situation: CVE-2026-86218 is confirmed exploited in the wild, unrelated to the two bugs patched a day earlier, and Huntress has documented an undocumented exploitation path.

N-able confirmed via its incident page that CVE-2026-86218, a max-severity unauthenticated RCE in N-central, has been observed exploited in the wild, and that it is separate from CVE-2026-86207 and the other flaw disclosed on 5 September. Hosted instances have vendor-applied mitigations; on-premises servers remain exposed until admins pull the upgrade from the N-able support portal. Huntress published details of an undocumented exploitation path, so MSPs should treat unpatched on-prem N-central as compromised and hunt rather than simply patch.

csoonline.com → (opens in a new tab)

Hackers exploit new MikroTik RouterOS flaws to hijack routers (opens in a new tab)

BleepingComputer ·Bill Toulas ·fetched 7 Sep 2026, 11:41 UTC Must read agreed3/3

Why readTwo chained RouterOS SSH bugs, CVE-2026-67276 (auth bypass via incomplete RSA public key validation) and CVE-2026-86060 (privilege escalation via crafted username), are being exploited against internet-exposed routers.

CVE-2026-67276 lets an attacker who knows a username and that user's public modulus craft a different key and authenticate without the legitimate private key; CVE-2026-86060 then escalates to full administrative privileges through a specially crafted SSH username. Poland's CERT found both, credits GPT-5.5-cyber and GPT-5.6-sol in the discovery, rated them critical, and named the chain MikroTrick after observing active exploitation. Anything with RouterOS SSH reachable from the internet needs the patch or the service taken off the perimeter now.

bleepingcomputer.com → (opens in a new tab)

N-able Issues Fourth N-central Hotfix in Five Weeks for Unauthenticated RCE Flaw (opens in a new tab)

The Hacker News ·The Hacker News ·fetched 7 Sep 2026, 11:41 UTC Must read agreed3/3

Why readCVE-2026-86218 is an unauthenticated RCE at CVSS 10.0 in N-able N-central, and every on-prem build below 2026.3.1.14 needs Hotfix 4, including servers patched to Hotfix 3 hours earlier.

N-able shipped its fourth N-central hotfix in five weeks, this one for a static code injection flaw (CWE-96) scored 10.0 under CVSS 4.0 by N-able as CNA, allowing code execution on the server without authentication. Build 2026.3.1.14 landed early on 6 September UTC and supersedes Hotfix 3 (2026.3.1.13), published roughly eight hours earlier for unrelated bugs. N-able's incident notice states in-the-wild exploitation while its own release notes call exploitation unconfirmed; RMM platforms are a standing initial-access target for ransomware crews, so treat the higher claim as operative.

thehackernews.com → (opens in a new tab)

Your MikroTik Router May Already Be Compromised: Look for SSH User “-2” (opens in a new tab)

Security Affairs ·Pierluigi Paganini ·fetched 7 Sep 2026, 07:38 UTC Must read CVE-2026-86060 EPSS 0.4% agreed3/3

Why readActive exploitation of internet facing MikroTik routers since 2 September, with patch versions and a one line indicator you can check right now.

The MikroTrick chain combines CVE-2026-67276, an SSH authentication bypass rated CVSS 9.2, with CVE-2026-86060, an SSH session privilege escalation, both from a set of six RouterOS flaws disclosed by CERT Polska. Costin Raiu published a technical breakdown on 5 September, the same day CERT Polska issued its advisory urging immediate updates. Patch to 7.24.2, 7.23.5 or 6.49.21, and look for an SSH user named "-2"; any device with SSH exposed should be treated as compromised until logs prove otherwise.

securityaffairs.com → (opens in a new tab)

Hackers Actively Exploiting PaperCut Servers Command Execution Vulnerabilities (opens in a new tab)

Cybersecurity News ·Abinaya ·fetched 7 Sep 2026, 07:38 UTC CVE-2026-81578 EPSS 1.7% agreed3/3

Why readCVE-2026-81578 and CVE-2026-82078 in PaperCut print management are being exploited for command execution and credential theft, with exploitation confirmed from 27 August and KEV listing on 31 August.

PaperCut has confirmed active exploitation of two command execution flaws beginning 27 August 2026, with CVEs assigned the following day and CISA adding both to the Known Exploited Vulnerabilities catalog on 31 August. Arctic Wolf Threat Intelligence reports attackers using compromised PaperCut servers as a foothold for credential theft and privileged account creation. PaperCut servers commonly sit inside enterprise networks with service account access to Active Directory, so a compromised print server is a domain problem rather than a printing one; the federal patch deadline applies and EPSS at 0.017 understates a bug already being used.

cybersecuritynews.com → (opens in a new tab)

CVE-2026-86218 (CVSS 10.0): N-central is vulnerable to a pre-auth remote code execution This issue affects N-central: before 2026.3.1.14. (opens in a new tab)

NVD ·fetched 7 Sep 2026, 19:38 UTC CVE-2026-86218 CVSS 10.0 EPSS 0.4% agreed3/3

Why readA pre-auth RCE at the top of the CVSS scale in an MSP remote monitoring platform, which is the class of software that turns one compromise into every downstream customer.

N-able N-central before 2026.3.1.14 carries an unauthenticated remote code execution flaw rated CVSS 10.0, with the vector showing network reach, no privileges, no user interaction and impact crossing the security boundary into managed systems. N-central sits above client estates with agent-level control, so exploitation is a tenant-wide problem rather than a single-server one. EPSS is still low at 0.4 percent, which reflects the absence of public exploit code so far and not the difficulty of the bug; treat the patch as urgent regardless.

Also covered byHelp Net Security (opens in a new tab),BleepingComputer (opens in a new tab),Infosecurity Magazine (opens in a new tab),Cybersecurity News (opens in a new tab).

nvd.nist.gov → (opens in a new tab)

Attackers spread malware through ScreenConnect file transfers (opens in a new tab)

Help Net Security ·Sinisa Markovic ·fetched 7 Sep 2026, 11:41 UTC agreed3/3

Why readConnectWise confirmed on 3 September that attackers are abusing a file transfer flaw in ScreenConnect support and access sessions, affecting both Cloud and On-Premise, with no CVE or fix yet.

A flaw in the file transfer function of ScreenConnect Remote Access Support and Access sessions is being used to deliver malware, per ConnectWise's 3 September advisory, which promises a CVE identifier and official fix within the week. Both ConnectWise-hosted and self-hosted deployments are affected. ScreenConnect is standard tooling for IT teams and MSPs and a long-standing favourite for attacker-delivered remote access, so hunt for unexpected file transfers in session logs while the patch is pending.

helpnetsecurity.com → (opens in a new tab)

CVE-2026-67276 (CVSS 9.2): RouterOS does not compare the complete RSA public key when matching an SSH authentication request to an authorized user key, checking the key type and (opens in a new tab)

NVD ·fetched 7 Sep 2026, 19:38 UTC CVE-2026-67276 CVSS 9.2 EPSS 0.2% agreed3/3

Why readRouterOS compares only key type and modulus when matching an SSH public key, so knowing an authorized RSA modulus (which is public) lets an attacker present exponent 1, forge a signature and log in without the private key.

The authorized-key match omits the exponent, and signature verification then uses the client-supplied key rather than the stored one. An attacker who knows a target user's RSA modulus supplies a key with exponent one, produces a signature that verifies trivially, and opens an SSH command channel as that user. Only the 7.x branch is affected; fixed in 7.23.4 (long-term) and 7.24.2 (stable). Audit which accounts have RSA keys authorised and where those moduli have been published.

nvd.nist.gov → (opens in a new tab)

ConnectWise warns of new ScreenConnect flaw without patch (opens in a new tab)

BleepingComputer ·Sergiu Gatlan ·fetched 7 Sep 2026, 11:41 UTC agreed3/3

Why readAn unpatched ScreenConnect file-transfer flaw with a concrete interim workaround: strip the TransferFiles (or legacy TransferFilesInSession) permission from session-group roles until the fix ships.

ConnectWise disclosed an issue in file transfer behaviour during ScreenConnect Remote Access Support and Access sessions, affecting both cloud and on-premises deployments, with no CVE assigned yet and a patch promised later in the week. The advisory's mitigation is to edit user roles, open Scoped Permissions for the affected session groups, and deselect TransferFiles or TransferFilesInSession. ScreenConnect is a standing target for ransomware affiliates through MSP access, so the workaround is worth applying before the patch lands.

bleepingcomputer.com → (opens in a new tab)

CVE-2026-67281 (CVSS 8.7): RouterOS WebFig contains an unauthenticated file-read vulnerability in the /jsproxy path where a newly allocated session retains a stale uninitialized (opens in a new tab)

NVD ·fetched 7 Sep 2026, 19:38 UTC CVE-2026-67281 CVSS 8.7 EPSS 0.5% agreed3/3

Why readUnauthenticated root-owned file read on MikroTik RouterOS 7.x WebFig, including config stores holding credentials; fixed in 7.23.4 and 7.24.2.

RouterOS WebFig allocates a new session that retains a stale uninitialized principal pointer used for file authorization on the /jsproxy path. An unauthenticated attacker can groom the allocator so the file-serving code dereferences that pointer with sufficient rights, then use parent-directory traversal inside an encrypted URI to escape the WebFig namespace and read root-owned files, including configuration stores containing credentials. Only the 7.x branch is affected; patched in 7.23.4 (long-term) and 7.24.2 (stable).

nvd.nist.gov → (opens in a new tab)

CVE-2026-86060 (CVSS 9.2): RouterOS contains an argument-handling flaw in the SSH login path involving usernames that begin with a prohibited character, allowing for the trusted (opens in a new tab)

NVD ·fetched 7 Sep 2026, 19:38 UTC CVE-2026-86060 CVSS 9.2 EPSS 0.4% agreed3/3

Why readAn unauthenticated SSH session to RouterOS can alter the trusted policy mask via a username starting with a prohibited character, escalating privilege on one of the most internet-exposed device classes there is.

RouterOS mishandles argument parsing in the SSH login helper when the username begins with a prohibited character, letting the trusted policy mask be changed and privilege escalated. Reaching the helper requires only an unauthenticated SSH session, so any router with SSH exposed is in scope. Fixed in 6.49.21 (long-term), 7.23.4 (long-term) and 7.24.2 (stable), the same release train as two other MikroTik bugs published today.

nvd.nist.gov → (opens in a new tab)

The NX bit is not just about security (opens in a new tab)

Hacker News ·torutofu ·fetched 7 Sep 2026, 07:38 UTC Research 62 points agreed3/3

Why readWalks a bare-metal ARM64 hypervisor lockup down to how the NX bit affects instruction fetch and cache behaviour, not just execution permission.

While building a hypervisor for postmarketOS, enabling the CTR_EL0 trap caused random lockups and watchdog resets on the target phone. The write-up follows the debugging from a suspected MRS emulation bug through to NX having consequences beyond blocking execution, on real hardware rather than in an emulator. Useful low-level ground truth for anyone doing ARM64 hypervisor, emulation or exploitation work where instruction fetch semantics matter.

purplesyringa.moe → (opens in a new tab) Discussion (opens in a new tab)

Conformal Prediction for Offensive Security (opens in a new tab)

arXiv cs.CR (all) ·Giovanni Cherubin ·fetched 7 Sep 2026, 19:38 UTC Research agreed3/3

Why readApplies conformal prediction to the attacker's side of privacy-preserving ML and network traffic analysis, giving calibrated confidence to membership and traffic-classification attacks rather than to defences.

Conformal prediction has been used almost exclusively defensively in security work; this paper takes it offensive, presenting initial results in two areas: attacks against privacy-preserving machine learning, and network traffic analysis. The framing is that CP's distribution-free coverage guarantees let an attacker quantify how much to trust a given inference, which matters for attacks whose value depends on precision. Explicitly preliminary findings rather than a finished technique, so treat it as a direction worth tracking.

arxiv.org → (opens in a new tab)

Has anybody seen my keys? A key-hierarchy strategy for rack-level security (opens in a new tab)

Hacker News ·cyb0rg0 ·fetched 7 Sep 2026, 15:40 UTC Research 47 points agreed3/3

Why readDesign detail on how a rack-level trust quorum built on Shamir secret sharing stops an attacker who physically walks off with a subset of sleds or drives.

Oxide's RFD 301 lays out the key hierarchy inside a rack: RoT-held DeviceId and Alias keys for platform identity and attestation signing, a third RoT keypair authenticating ephemeral Diffie-Hellman for sprockets sessions between sleds, and above that a rack-level secret split with Shamir secret sharing to form a trust quorum. The threat model is explicit and physical: recovering useful data must require a quorum of hardware, not any single stolen sled or disk. Useful as a worked reference for anyone designing platform key hierarchies or evaluating attestation claims from hardware vendors.

rfd.shared.oxide.computer → (opens in a new tab) Discussion (opens in a new tab)

The History Is the Detector: Executing CVE Patch History, End-to-End (opens in a new tab)

arXiv cs.CR (AI) ·Qiushi Wu, Kevin Eykholt, Youngja Park, Xiaokui Shu ·fetched 7 Sep 2026, 03:42 UTC Research agreed3/3

Why readTurns verified CVE fixing commits into executable detection rules that find the same unsafe pattern in code with no advisory of its own.

BUGSTONE-E2E mines reusable rules from fixing commits, capturing scan anchors, fix semantics and CVE provenance, then organises them by CWE and language. Detection runs as a funnel: cheap static analysis enumerates a large candidate pool, progressively more expensive models are applied to the shrinking set, and findings are validated rather than reported raw. The interesting claim for appsec teams is that patch history is an underused detection corpus, not just documentation for humans.

arxiv.org → (opens in a new tab)

Understanding the Privacy-Preserving Potential of HTTP/2 Against Webpage Fingerprinting (opens in a new tab)

arXiv cs.CR (all) ·Bogdan Cebere, Prateek Kumar, Sylvain Chatel, Wouter Lueks ·fetched 7 Sep 2026, 23:41 UTC Research agreed3/3

Why readWebsite fingerprinting defenses have mostly been studied inside Tor or a VPN; this shows the same protections can be built from HTTP/2 features that are already deployed at both endpoints.

The authors reimplement known fingerprinting defenses, including HTTPOS, LLaMA, FRONT, Tamaraw and ALPaCA, using ordinary HTTP/2 mechanisms on the client and the server, then go further and build lightweight defenses out of proactive resource suggestion, multiplexing and flow control. Everything is evaluated through one blueprint that tunes parameters per dataset and reports practical attack accuracy, information theoretic leakage and bandwidth or latency overhead side by side. The result is a realistic picture of what application layer traffic shaping buys you without an encapsulating protocol, which matters for anyone weighing privacy protections they can actually ship on a web property.

arxiv.org → (opens in a new tab)

Propagation Model for SSC attacks: Why SBOM (tools) don't tell the whole truth (opens in a new tab)

arXiv cs.CR (all) ·Ljubica Grgic, Lazar Maksimovic, Pavel Laskov ·fetched 7 Sep 2026, 03:42 UTC Research agreed3/3

Why readTests four open-source SBOM tools against Log4j and finds none of them reach code reachability or taint analysis, only structural exposure and vulnerability class presence.

The authors define a four-stage propagation model for software supply chain risk and evaluate four open-source SBOM tools across three projects using Log4Shell as the test case. Tools consistently handle Stage 1 structural exposure and Stage 2 vulnerability class presence, while Stage 3 code reachability and Stage 4 taint path analysis require capabilities the SBOM ecosystem does not have. The practical conclusion is that an SBOM-derived vulnerability list tells you a component is present, not that it is exploitable, which is the gap teams keep mistaking for a finding.

arxiv.org → (opens in a new tab)

Repeat-After-Me: Black-Box Adaptive Visual Prompt Injection (opens in a new tab)

arXiv cs.CR (AI) ·Sizhe Chen, Yu-Lin Tsai, Ivan Evtimov, Kamalika Chaudhuri ·fetched 7 Sep 2026, 07:38 UTC Must read Research agreed3/3

Why readBlack-box image prompt injection that reaches 80% attack success on Qwen3.6-27B and 47% on GPT-5.5, including well-formed malicious tool calls.

Repeat-After-Me is an adaptive black-box attack that solves the hard part of visual prompt injection: emitting long, format-compliant target strings such as a parseable native tool call with exact function names and arguments. Tested against open-weight and frontier commercial VLMs, it exfiltrates PII and triggers malicious tool calls at over 80% and 47% success respectively, under the realistic condition that the user's own prompt is unrelated to the injected task and never authorizes it. That closes much of the gap between text and image injection, so any agent pipeline that lets a model read untrusted screenshots or attachments now needs the same distrust applied to pixels as to text.

arxiv.org → (opens in a new tab)

When LLM Decompilers Recompile More and Preserve Less (opens in a new tab)

arXiv cs.CR (AI) ·Chang Liu, Edward Raff, Kristopher Micinski ·fetched 7 Sep 2026, 03:42 UTC Must read Research agreed3/3

Why readEmpirical evidence that the two metrics everyone uses to judge LLM decompilers, recompilability and re-executability, can certify output that has silently deleted the vulnerability you were trying to analyse.

The authors show that LLM decompilers produce clean idiomatic C that builds and passes its shipped input/output tests while diverging from the original binary on other legitimate inputs, and that a disclosed vulnerability can vanish from the recompiled code leaving no placeholder or artifact to signal the loss. Traditional decompilers like Ghidra and Hex-Rays at least surface what they cannot resolve; the LLM output looks correct precisely where it is wrong. Their Decompile-Diverge oracle synthesises a driver per function, grows a fuzzing corpus from the reference binary, and replays the same inputs against the decompiled version to surface behavioural divergence that fixed test suites miss.

arxiv.org → (opens in a new tab)

Machine Unlearning as Private Retroactive Algorithms (opens in a new tab)

arXiv cs.CR (all) ·Haim Kaplan, Refael Kohen, Yishay Mansour, Kobbi Nissim ·fetched 7 Sep 2026, 11:41 UTC Research agreed3/3

Why readArgues machine unlearning provides no privacy guarantee against an adversary watching a sequence of releases, and replaces it with private retroactive algorithms achieving differential privacy under continual observation at no asymptotic cost for linear statistics, clustering and histograms.

Reframes unlearning as a data maintenance problem rather than a privacy one: emulating retraining from scratch carries no meaningful privacy semantics once an adversary sees successive model releases. The authors define private retroactive algorithms, which combine retroactivity (all later answers reflect the revised history as if it had always held) with differential privacy under continual observation, and give constructions plus impossibility results. Directly relevant to anyone treating a deletion request pipeline as a compliance answer for GDPR erasure against a deployed model.

arxiv.org → (opens in a new tab)

Top MCP security resources — September 2026 (opens in a new tab)

Adversa AI ·fetched 7 Sep 2026, 07:38 UTC agreed3/3

Why readPopulation level numbers on how internet facing MCP servers are actually deployed, and they are worse than the individual bug reports suggest.

The headline resource scanned eleven sources to find MCP servers, then ran 34 test modules covering ten MCP specific vulnerability classes against 414 of 640 confirmed production servers, surfacing 68 reportable issues including SQL injection, SSRF against cloud metadata, prompt template injection and path traversal through cursor manipulation. Posture was the worse finding: 91.8 percent had no authentication at all and 687 tool instances exposed shell execution with no access control, while 41.6 percent of servers disappeared within three days. A second item covers a Confluence attachment tool passing a client supplied path straight to open(), giving any authenticated client an arbitrary file read across the server process environment.

adversa.ai → (opens in a new tab)

AI Coding Agents Are Installing Unknown/Untrusted Code on Corporate Networks (opens in a new tab)

Schneier on Security ·Bruce Schneier ·fetched 7 Sep 2026, 11:41 UTC Must read agreed3/3

Why readUnregistered packages and domains referenced in public llms.txt files get pulled and executed by coding agents, and the researchers proved it with a phone-home beacon.

A scan of 6,214 live domains belonging to defense contractors, Fortune 500 firms and Big Tech found 8,265 llms.txt and llms-full.txt files, of which 120 (each on a different site) referenced code packages or domain names that were never registered. The researchers registered several of the unclaimed names, hosted packages that beacon on execution, and had a callback from a Fortune 500 company within an hour, with parent-process chains implicating Claude, OpenAI Codex and Nous Research's Hermes. This is dependency confusion moved into agent-readable context files, and it is a commentary write-up rather than the primary report.

schneier.com → (opens in a new tab)

Privacy Failure in Split-LLM Training, The Returned Gradient Nullifies the Decoys (opens in a new tab)

arXiv cs.CR (AI) ·Georgios Politis, Evangelos Pappas ·fetched 7 Sep 2026, 07:38 UTC Research agreed3/3

Why readA split-LLM training scheme that passed its own privacy evaluation leaks which rows are real, because decoy gradients return as exact zeros.

In the two-node design examined, the trusted local node mixes real rows with decoys before sending activations to the untrusted cloud node, but the loss ignores decoys, so the returned output gradient carries exactly zero for every decoy row. Under a pre-registered protocol with an injected known-strength leak, a shuffled-label control and a threshold fixed before the runs, the zero pattern identified all 4,096 real rows on every frame across nine seeds, and content recovery beat a constant-guess baseline by 0.65 to 1.50 percentage points. The wider lesson for anyone reviewing confidential-compute or split-inference claims is that the return channel is part of the attack surface and is routinely left out of the evaluation.

arxiv.org → (opens in a new tab)

CONTINUITY: Security-Context Contracts for Composable LLM Agent Controls (opens in a new tab)

arXiv cs.CR (AI) ·Chris Zheng, Geng Yang ·fetched 7 Sep 2026, 03:42 UTC Research agreed3/3

Why readNames a concrete failure mode in agent security stacks: individually correct provenance, authz and policy components that drop or widen security context at the boundaries between them.

CONTINUITY models each component of an LLM agent stack with an assume-guarantee contract and carries authenticated context across transitions using signed root grants, provenance commitments, role-bound transition receipts, bounded typed releases and effect-bound execution permits. The property it enforces, end-to-end consequence integrity, requires every external effect to trace back to a current authorization witness binding principal, task, provenance, delegation and canonical action. A reference implementation exists, though the work is formal rather than an attack against deployed systems.

arxiv.org → (opens in a new tab)

GPT-6 Astra Scores 100% on ExploitBench as OpenAI Blocks PoC Exploit Requests (opens in a new tab)

The Hacker News ·The Hacker News ·fetched 7 Sep 2026, 11:41 UTC Must read agreed2/3

Why readA frontier lab has now declared a model past the Critical cybersecurity threshold, and its stated control is refusing to write proof-of-concept exploits.

OpenAI launched GPT-6 Astra claiming a 100 percent score on ExploitBench, which measures turning known vulnerabilities into working exploits, alongside a Critical cyber capability rating under its own Preparedness Framework. The benchmark figures are vendor-reported and unverified, so treat them as a claim rather than a measurement; what is not in question is the posture change, since the declared mitigation is blocking requests for PoC exploit code. The planning assumption for defenders is that the interval between public disclosure and a working exploit keeps compressing for anyone with API access and a way around the refusal.

thehackernews.com → (opens in a new tab)

Rethinking Indirect Prompt Injection as a Test-Time Search Problem (opens in a new tab)

arXiv cs.CR (AI) ·Duong M. Nguyen, Joon Sik Kim, Blazej Manczak, Vaikkunth Mugunthan ·fetched 7 Sep 2026, 07:38 UTC Research agreed3/3

Why readReframes indirect prompt injection as attacker-side search, and shows attack success scales with the attacker's test-time compute rather than being a fixed property of the victim agent.

The authors build an agentic attacker with a search harness that performs environment reconnaissance, reasons over candidate injection strategies, and adapts using feedback from the victim agent. More attacker compute yields more discovered and exploited vulnerabilities, and ablations show explicit strategy management is what prevents redundant search from flattening the gains at larger budgets. The practical consequence is that a red-team result of "our agent resisted injection" is meaningless without stating the attacker's search procedure and compute budget.

arxiv.org → (opens in a new tab)

Forgetting Without Restarting: Execution-State Unlearning for Stateful LLM Agents (opens in a new tab)

arXiv cs.CR (AI) ·Chao Yao, Yangbo Wei, Zhen Huang, Junhong Qian ·fetched 7 Sep 2026, 03:42 UTC Research agreed3/3

Why readExplains why a 'forget' on a long-running agent is mostly theatre: deleting the memory record leaves summaries, pending tool plans and the KV cache still tainted.

The paper formalises execution-state unlearning, requiring an agent to behave as if it had never observed the revoked data, and proves that the tainted suffix cannot be repaired without token-level attribution and that exact unlearning needs at least T minus tau plus one recomputed transitions from the injection step. Provenance-Guided Selective Replay hits that bound by locating the injection point in a provenance graph, cropping the KV cache back to a checkpoint, and replaying a sanitised suffix. Relevant to anyone handling deletion requests or credential revocation in stateful agent deployments.

arxiv.org → (opens in a new tab)

CVE-2026-86121 (CVSS 9.3): Cua computer-server versions before 0.3.42 skip authentication when the CONTAINER_NAME environment variable is unset and bind to all interfaces by def (opens in a new tab)

NVD ·fetched 7 Sep 2026, 23:41 UTC CVE-2026-86121 CVSS 9.3 EPSS 0.6% agreed3/3

Why readA computer use agent sandbox that silently drops authentication when a single environment variable is missing, while listening on every interface by default.

Cua computer-server before 0.3.42 skips authentication entirely if CONTAINER_NAME is unset, and binds to all interfaces out of the box. Anyone able to reach TCP 8000 gets the run_command endpoint, arbitrary file read and write through the file operation endpoints, and interactive PTY shells. The pattern is worth checking across your own agent infrastructure: absent config resolving to absent auth, rather than a refusal to start.

nvd.nist.gov → (opens in a new tab)

CVE-2026-86169 (CVSS 8.7): Axolotl through 0.18.0 contains a remote code execution vulnerability in the multipack patch path where trust_remote_code defaults to None instead of (opens in a new tab)

NVD ·fetched 7 Sep 2026, 19:38 UTC CVE-2026-86169 CVSS 8.7 EPSS 0.5% agreed3/3

Why readA worked example of how a security guard gets silently disabled: trust_remote_code defaults to None rather than False, so the falsy check passes and the loader hardcodes True anyway.

Axolotl through 0.18.0 executes arbitrary Python from a malicious Hugging Face repository chosen as base_model, because the multipack patch path initialises trust_remote_code to None instead of False and the guard that should block remote code never fires; AutoModelForCausalLM.from_pretrained is then called with trust_remote_code hardcoded to True. The practical exposure is any fine-tuning pipeline that takes a model identifier from a config file, a ticket or an untrusted contributor. It is worth auditing your own loaders for the same None-versus-False pattern, since the bug is in the default, not the logic.

nvd.nist.gov → (opens in a new tab)

Japan's Health Ministry to Strengthen Cybersecurity Measures at Hospitals (opens in a new tab)

translated Japan’s Health Ministry to Strengthen Cybersecurity Measures at Hospitals

DataBreaches.net ·Dissent ·fetched 7 Sep 2026, 15:40 UTC agreed3/3

Why readA national health regulator putting real money behind hospital network defence and staffing, useful as a benchmark when arguing for sector funding elsewhere.

Japan's Health, Labour and Welfare Ministry has included ¥13.7 billion in its fiscal 2027 budget request to harden cybersecurity at medical institutions, covering network protection and the deployment of security specialists into hospitals. The move responds to a sustained rise in attacks on Japanese healthcare providers, where ransomware has repeatedly taken clinical systems offline. It is a budget request rather than an approved allocation, so the figure is a statement of policy direction more than committed spend.

databreaches.net → (opens in a new tab)

NCSC Warns Shadow AI Creates New Security Risks (opens in a new tab)

Infosecurity Magazine ·fetched 7 Sep 2026, 23:41 UTC agreed2/3

Why readNCSC has now put a national position on shadow AI in writing, and the 71 percent adoption figure it cites is the number your board will quote back at you.

The UK NCSC published guidance arguing that unapproved employee AI use will persist because staff adopt services faster than organisations can assess them and offer sanctioned equivalents. It cites Microsoft research finding 71 percent of UK employees have used AI tools their employer did not approve, and warns that security teams should not assume monitoring gives them the full picture, particularly where employees connect these tools to corporate data. The content is positional rather than technical, so its use is in policy and vendor assessment conversations rather than detection engineering.

infosecurity-magazine.com → (opens in a new tab)

CREST Onboards First Cohort for AI-Enabled Pentesting Accreditation (opens in a new tab)

Infosecurity Magazine ·fetched 7 Sep 2026, 03:42 UTC agreed2/3

Why readThe first accreditation scheme that formally assesses how testing providers use AI, which is the answer procurement teams have started asking suppliers for.

CREST has accredited an initial cohort of 10 providers across Europe, India and the US under an AI-Enabled Penetration Testing module added to its Penetration Testing Accreditation Standard in July 2026. The module is optional and does not affect standard membership; it exists so firms already using AI in delivery can demonstrate governance to clients and regulators through independent assessment. Useful mainly as a procurement signal, since it establishes a checkable credential where previously buyers had only vendor self-assertion.

infosecurity-magazine.com → (opens in a new tab)

EFF to Courts: Don’t Rewrite Copyright Over AI Hype (opens in a new tab)

EFF Deeplinks ·Corynne McSherry ·fetched 7 Sep 2026, 11:41 UTC agreed2/3

Why readLays out the doctrinal line, substantial non-infringing use from the Betamax case, that courts are now being asked to apply to model training.

EFF argues that judges hearing AI copyright suits should resist calls to expand liability, drawing a direct parallel to the 1984 Sony decision that spared the videocassette recorder because it had legitimate uses. The piece stacks up earlier copyright panics, the player piano, the gramophone, the camera, to argue the predicted collapse of creative markets never arrived. It is advocacy rather than reporting, useful mainly as a map of the arguments defendants in training-data litigation will lean on.

eff.org → (opens in a new tab)

Offensive Security Best Practices for Modern Enterprises (opens in a new tab)

IOActive ·Christian Powills ·fetched 7 Sep 2026, 07:38 UTC agreed2/3

Why readA programme design for continuous offensive testing that commits to a specific starting cadence rather than the usual maturity model abstraction.

IOActive argues the annual penetration test is insufficient and lays out a threat informed alternative: pick two objectives per cycle, one path to a high value asset and one to a critical service, map a single chained attack path, and define what evidence must exist at each step. It then sequences an independent red team exercise followed by a purple team retest of the highest risk gap. The audience is security leadership building the capability rather than operators executing it, and the guidance stays at programme level throughout.

ioactive.com → (opens in a new tab)

Jaguar Land Rover to Slash 4,000 Jobs, Says Cyberattack to Blame (opens in a new tab)

Google News: incidents · Autoguide.com ·fetched 7 Sep 2026, 23:41 UTC Must read agreed3/3

Why readJaguar Land Rover is cutting 4,000 jobs and naming last year's cyberattack as the cause, the clearest public example yet of an incident translating directly into headcount.

JLR is reducing its workforce by 4,000 and attributes the decision in part to the cyberattack that halted its production lines. It is the kind of downstream figure a board will cite when asking what an outage of that length would cost here. Manufacturing peers should expect the question this week.

Also covered byDaily Sabah (opens in a new tab).

news.google.com → (opens in a new tab)

Hackers drain $320M in Bitcoin from Liquid Network, claim they’re the good guys (opens in a new tab)

DataBreaches.net ·Dissent ·fetched 7 Sep 2026, 19:38 UTC agreed3/3

Why readRoughly 4,000 BTC, about $320M, drained from the Liquid Network federation wallet, with the attackers framing it as a white-hat action.

Blockstream's Liquid sidechain confirmed on X that around 4,000 BTC left the federation multisig wallet backing the network, worth roughly $320 million, in an incident the attackers publicly claim was benign. Liquid is used by exchanges and other financial institutions, so the exposure question reaches counterparties rather than just Blockstream. The write-up is a short aggregation of TechCrunch reporting with no detail yet on how federation signing was subverted.

databreaches.net → (opens in a new tab)

US and Canadian Court Records Breached Following Thomson Reuters Incident (opens in a new tab)

Infosecurity Magazine ·fetched 7 Sep 2026, 07:38 UTC Must read agreed3/3

Why readThomson Reuters has disclosed that an intruder took court records from its C-Track case management system, including sealed and redacted material from three Ontario courts.

Activity affecting C-Track, Thomson Reuters' digital case management product, was detected on 30 June, and the investigation found an unauthorised party obtained C-Track Canada files tied to the Court of Appeal for Ontario, the Ontario Superior Court of Justice and the Ontario Court of Justice. Affected records may contain names and personal information, and the company acknowledges that confidential, redacted or sealed information may have been exposed for some courts. Any organisation whose litigation records sit in a third-party court platform now has a concrete precedent to raise with counsel, and the exposure of sealed material makes this more than a routine PII notification.

infosecurity-magazine.com → (opens in a new tab)

Risky Bulletin: BEC campaign steals €35 million from French notaries (opens in a new tab)

Risky Business News ·fetched 7 Sep 2026, 07:38 UTC agreed3/3

Why readA BEC campaign took 35 million euros from French notaries, and Five Eyes agencies are now telling breached companies to cut the PR spin from disclosures.

Attackers extracted 35 million euros from French notaries via business email compromise, a sector whose escrow and conveyancing accounts make it a standing target for payment-diversion fraud. The same bulletin covers Five Eyes guidance pressing breached organisations toward plainer incident disclosure, a US bill letting the Pentagon contract offensive cyber work, and OpenAI agents compromising a German wiki. The notary losses and the disclosure guidance are both things a board or general counsel will ask about.

risky.biz → (opens in a new tab)

Data Breach at American Clothing Giant Carhartt Exposes Nearly 13 Million People (opens in a new tab)

Google News: incidents · CPO Magazine ·fetched 7 Sep 2026, 11:41 UTC agreed3/3

Why readCarhartt has disclosed a breach affecting close to 13 million people, a retail incident at consumer scale that peers and boards will be asked about.

Reporting via CPO Magazine puts the exposure at nearly 13 million individuals at the US clothing manufacturer Carhartt. The item reaches us as an aggregator headline with no detail on data categories, intrusion vector or notification timeline, so the population figure and the named company are all that is confirmed. Worth flagging for anyone benchmarking retail breach scale or watching for the state attorney general notifications that a number this size will generate.

news.google.com → (opens in a new tab)

Grindr to pay £26m to settle claims it allegedly shared users' HIV status (opens in a new tab)

BBC Technology ·fetched 7 Sep 2026, 15:40 UTC agreed3/3

Why readA £26m privacy settlement over sharing HIV status data, disclosed via SEC filing, sets a price tag on legacy data-sharing practices that boards will ask about.

Grindr agreed on 2 September to pay £26m, in two £13m instalments, to settle a UK High Court class action alleging it shared users' personal information including HIV status with third parties; the first payment is due by 31 December. The law firm leading the case signed up more than 11,000 claimants, and the settlement was disclosed in a US SEC filing with no admission of liability. Grindr attributes the conduct to "historical data practices" before 2020 under prior owner Kunlun, which is the argument any organisation inheriting an acquired product's data pipeline will end up making.

Also covered byLaw360 (opens in a new tab).

bbc.co.uk → (opens in a new tab)

Trezor data breach impact now reaches 81,000 customers (opens in a new tab)

BleepingComputer ·Sergiu Gatlan ·fetched 7 Sep 2026, 15:40 UTC agreed3/3

Why readA third-party logistics provider kept customer data it was contractually required to delete, expanding a hardware wallet breach from 14,000 to 81,000 people.

Trezor's August breach at fulfilment provider ShipMonk now covers an additional 67,000 US customers who ordered between November 2019 and August 2021, with names, emails, phone numbers, shipping addresses and order numbers exposed. Trezor says ShipMonk failed to delete the data as its contract and data policy required, having repeatedly been asked to do so. The retention failure, not an exploit, is the story: physical shipping addresses tied to crypto hardware buyers are a targeting list, and the vendor-obligation gap is the part a board will ask about.

bleepingcomputer.com → (opens in a new tab)

Berlin investigates new data leak after hackers publish stolen login credentials (opens in a new tab)

The Record ·fetched 7 Sep 2026, 15:40 UTC agreed3/3

Why readA second data release from the August compromise of two Berlin state ministries now includes login credentials, with the city unable to say whether they are still valid.

Attackers published a fresh tranche of data over the weekend from the mid-August breach of Berlin's ministries for urban development and housing and for transport, mobility, climate protection and the environment. The city confirmed the leak contains login credentials but would not say what they access or whether they remain live, and no attribution has been made. Berlin's data protection authority says a large volume of data was taken; the urban development ministry has tightened controls in ways that may curtail access to some of its own applications.

therecord.media → (opens in a new tab)

7th September – Threat Intelligence Report (opens in a new tab)

Check Point Research ·urias ·fetched 7 Sep 2026, 15:40 UTC agreed2/3

Why readThree fresh breach disclosures with scope attached, including a court case-management platform compromised across 11 US states and Canada.

Check Point's weekly bulletin reports that Thomson Reuters disclosed unauthorized access to C-Track files containing court records from courts in 11 US states and Canada, and that Baylor Genetics notified 2.8 million patients and employees after an intrusion in June exposed names, birth dates, lab results and insurance details. Slovenian operator Hit lost about three days of casino operations to an attack, with table games, bingo, loyalty and hotel systems degraded through the restoration. The value here is the named victims and the record counts, not the analysis; treat it as a disclosure log rather than research.

research.checkpoint.com → (opens in a new tab)

MINNEAPOLIMEDIA NEWS | Winona County Paid $128,539 Ransom After Cyberattack, Then Was Attacked Again Three Months Later (opens in a new tab)

Google News: incidents · MinneapoliMedia ·fetched 7 Sep 2026, 15:40 UTC agreed3/3

Why readWinona County paid a $128,539 ransom and was hit again three months later, a concrete data point for anyone arguing the payment question upward.

The Minnesota county disclosed paying $128,539 after a ransomware incident, then suffered a second attack roughly three months on. The named figure and the repeat compromise make this usable evidence that payment buys neither eviction of the intruder nor durable protection. Local-government reporting rather than technical analysis; no actor or malware family named.

news.google.com → (opens in a new tab)

HYBE's Weverse Platform Suffers Data Breach Affecting 420,000 Users, Exposing Payment Records (opens in a new tab)

Google News: incidents · finance.biggo.com ·fetched 7 Sep 2026, 03:42 UTC agreed3/3

Why readNamed breach at HYBE's Weverse fan platform affecting 420,000 users with payment records exposed, which puts consumer notification and regulator questions in play.

Weverse, the fan engagement platform operated by HYBE, disclosed a breach affecting roughly 420,000 users, with payment records among the exposed data. The item arrived as a headline with no technical detail on intrusion vector or timeline. The user count and the payment data element are the facts that matter for anyone assessing consumer-platform exposure or Korean regulatory response.

Also covered byKorea JoongAng Daily (opens in a new tab),Bitdefender (opens in a new tab).

news.google.com → (opens in a new tab)

DentaQuest sued for allegedly exposing 15M patients’ private information (opens in a new tab)

Google News: incidents · Top Class Actions ·fetched 7 Sep 2026, 15:40 UTC agreed3/3

Why readA class action over a claimed 15 million-patient exposure at DentaQuest, the litigation tail that follows a healthcare breach.

DentaQuest faces a proposed class action alleging it exposed the private information of roughly 15 million dental patients. The filing itself is the news; the complaint's allegations are untested and the item carries no technical detail on how the data was reached. Useful as a cost-of-breach reference point for healthcare and insurance risk owners.

news.google.com → (opens in a new tab)