Thursday, 5 November 2026

Binance says hackers who stole billions in crypto spent months planning heist (opens in a new tab)

Ars Technica Security ·Dan Goodin ·fetched 5 Nov 2026, 03:41 UTC Must read agreed3/3

Why readThe victim's own timeline for the largest crypto theft on record, including how long the operators were inside and what finally gave them the keys.

Binance says the actors behind the theft of $6.34 billion in cryptocurrency began work no later than June and ran until the compromise of an engineer's laptop last Thursday allowed withdrawals from several wallets. The operation combined social engineering with malware the company describes as novel, and both the FBI and TRM attribute it to North Korean actors. Binance has frozen or intercepted $2.85 billion and says it has reserved funds to cover the remainder.

arstechnica.com → (opens in a new tab)

Hacking on Autopilot: Inside ARTEX Cybercrime Campaigns (opens in a new tab)

Check Point Research ·stcpresearch ·fetched 5 Nov 2026, 19:41 UTC Must read Research agreed3/3

Why readRare look at the attacker side of an agentic hacking platform, built from operator console artefacts rather than victim telemetry.

Check Point counted more than 1,800 publicly exposed ARTEX instances and pulled artefacts from the operator consoles themselves, revealing which models criminals actually run and how they prompt their agent fleets. Chinese models dominate, with Deepseek Flash in the lead. Operators set a goal and then largely walked away, checking progress rather than steering; the objectives observed were uniformly opportunistic, financially motivated cybercrime at scale.

Also covered byThe Hacker News (opens in a new tab).

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

The SOC Files 5: Ghost in the pipeline, or a fileless SharePoint breach by Storm-2603 (opens in a new tab)

Securelist ·Alexander Rodchenko, Rodrigo Wanderson ·fetched 5 Nov 2026, 11:38 UTC Must read Research CVE-2025-53770 EPSS 99.9% agreed3/3

Why readWalks a Storm-2603 ToolShell intrusion where the attacker kept the whole access layer inside w3wp.exe, writing nothing to disk for persistence.

Retrospective analysis of a customer intrusion traced to Storm-2603 exploiting CVE-2025-53770 and CVE-2025-53771 in on-premises SharePoint for RCE via crafted HTTP requests. The attacker used the IIS worker process as an in-memory bridge into the network, so the command and control layer left no disk artefacts even though later post-exploitation stages did. CVE-2025-53770 sits at EPSS 0.9989, so anyone still running unpatched on-prem SharePoint should treat the memory-resident tradecraft described here as the hunting baseline rather than the file-based IOCs.

securelist.com → (opens in a new tab)

CWPdoor: The Invisible Backdoor Living Inside Control Web Panel (opens in a new tab)

QiAnXin XLab ·Alex.Turing ·fetched 5 Nov 2026, 15:39 UTC Research agreed3/3

Why readFull analysis of cwpdoor, a passive backdoor compiled into the Control Web Panel cwpsrv binary that triggers on crafted HTTP headers and suppresses its own access log entries.

XLab's CTIA caught a script named cwpsrv_update.sh on 29 June 2026 pulling a replacement cwpsrv binary from 136.0.141.71. The attacker patched malicious code into the cwpsrv source and recompiled, so the server behaves normally until a specific header triggers remote command execution, and the request is then kept out of the access logs. Anyone running Control Web Panel should hash-verify cwpsrv against a known-good build rather than trusting process behaviour or logs.

Also covered byQiAnXin XLab (opens in a new tab).

blog.xlab.qianxin.com → (opens in a new tab)

SlippyNunchucks hijacks Sogou Input Method’s brand promotion mechanism to distribute malware (opens in a new tab)

ESET WeLiveSecurity ·fetched 5 Nov 2026, 03:41 UTC Research agreed3/3

Why readShows how SlippyNunchucks abused the plaintext HTTP brand-promotion channel in Sogou Input Method, via DNS hijacking, to deliver DarkVigil implants.

ESET resolves a question it had left open since 2025: how the SlippyNunchucks actor gets DarkVigil (also tracked as FDMTP) onto machines through popular Chinese applications. The vector is the Sogou Input Method for Windows fetching brand-promotion content over insecure HTTP, which an adversary in a position to hijack DNS can replace with malware. The shift followed the QuickFox supply-chain compromise, with DarkVigil deliveries continuing from 1 July 2026, three days after QuickFox shipped a clean build, giving defenders a concrete second delivery path to hunt for in Chinese-speaking user populations in Asia and Europe.

welivesecurity.com → (opens in a new tab)

Want to Rent an Internet Address at the White House? (opens in a new tab)

Krebs on Security ·BrianKrebs ·fetched 5 Nov 2026, 07:39 UTC agreed3/3

Why readResidential proxy software was running on Executive Office of the President endpoints, renting out White House egress to anyone who paid.

Layer3 Intel found proxy nodes beaconing from AS2650, the network reserved for the Executive Office of the President, persistently between 5 October and 3 November, with 236 recorded exits through that space. Synthient attributed the activity to Infatica installations on multiple Windows 11 machines; Infatica is commonly bundled into free mobile and desktop apps, and its founder previously ran the hosting operation tied to the 2016 attacks on Arizona and Illinois election systems. The practical lesson is that proxy SDKs reach sensitive networks through ordinary consumer software rather than through targeted intrusion, and endpoint inventories rarely flag them.

krebsonsecurity.com → (opens in a new tab)

APT29 Watering Holes Target Unpatched iPhones and Android Devices (opens in a new tab)

Infosecurity Magazine ·fetched 5 Nov 2026, 03:41 UTC agreed3/3

Why readSekoia's attribution of a mid-October 2026 mobile watering-hole campaign to APT29, including a week-long compromise of the Ukrainian analytics tracker Plerdy as a delivery route.

Scripts injected into legitimate sites, lookalike domains proxying real sites, and the Plerdy tracker compromise served exploit chains for already-patched iOS and Android flaws, with payloads stealing cookies, device data and Wi-Fi passwords without user interaction. Sekoia assesses with moderate to high confidence that APT29 (Midnight Blizzard, Cozy Bear) or a subgroup is responsible; targets included a Ukrainian gas producer and Ukrainian military and military-equipment sites. This is Infosecurity's write-up rather than the Sekoia research itself, so go to the primary post for indicators, but the delivery-chain detail is enough to start checking mobile fleet patch levels and analytics-script supply chain.

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

Malicious browser extensions ecosystem overview (opens in a new tab)

Intrinsec ·Ruben Madar ·fetched 5 Nov 2026, 19:41 UTC Research agreed3/3

Why readDocuments the malicious browser extension lifecycle, clean extension first then a silent malicious update, and gives infrastructure pivots plus panel IPs for the Needle and Lunex extension malware families.

Intrinsec maps recent malicious extension campaigns and the shared playbook behind them: publish a clean extension to build trust and install base, push the payload in a silent update, then evade analysis while stealing cryptocurrency wallets and authenticated sessions. Underground forum discussion is quoted on the demand for silent installation and on store moderation as a stability problem for operators. The paper includes a SOC and CTI incident involving suspicious extensions, and infrastructure analysis identifying IP addresses hosting Needle and Lunex panels.

intrinsec.com → (opens in a new tab)

Residential Proxy Traffic Traced to White House Network (opens in a new tab)

Infosecurity Magazine ·fetched 5 Nov 2026, 23:41 UTC agreed3/3

Why readConcrete, independently corroborated evidence that commercial residential proxy networks are selling exits inside the Executive Office of the President's address space.

Layer3 Intel recorded 236 residential proxy exits from six addresses in AS2650, the block registered to the Executive Office of the President, on 26 of the 30 days between October 5 and November 3, with 28 distinct proxy services able to route through them. The firm cross-checked the addresses against public lookups from Spur and Synthient, both of which flagged them independently, which raises the finding above a single vendor's telemetry. The practical lesson generalises past this one network: proxy SDKs bundled into consumer apps turn any managed device into a sellable exit node, and your egress ranges inherit whatever those customers do.

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

Google, Partners Disrupt 1.3 Million-Device-Strong XCPE Botnet (opens in a new tab)

Dark Reading ·Elizabeth Montalbano ·fetched 5 Nov 2026, 03:41 UTC agreed3/3

Why readGoogle and partners disrupted XCPE, a 1.3 million device botnet of backdoored Android set-top boxes and IoT devices, with Google naming manufacturer Videostrong as the operator.

The botnet was built from set-top boxes and IoT devices shipped with a backdoor, giving operators criminal infrastructure at a scale of roughly 1.3 million devices. Google's allegation that the device manufacturer Videostrong ran the operation is the sharp end of this: it is a supply-chain compromise at the point of manufacture rather than post-sale infection, which is a different problem for anyone with cheap Android hardware on their network. This is Dark Reading's write-up; Google's own disruption report will carry the indicators and the sinkhole detail.

darkreading.com → (opens in a new tab)

Ukraine fires two senior security officials amid corruption probe into scam call centers (opens in a new tab)

The Record ·fetched 5 Nov 2026, 11:38 UTC agreed3/3

Why readConcrete evidence of how scam call centres buy protection: NABU and SAPO allege a bribery network reaching senior SBU officials, two of whom have now been fired.

Ukraine's security service dismissed two senior officials after NABU and SAPO searched properties tied to several SBU figures as part of Operation Carthage, an investigation into a criminal organisation of at least twelve people that allegedly paid bribes so fraudulent call centres could run unimpeded. SBU chief Oleksandr Poklad announced the dismissals the same evening the searches became public. For anyone modelling the fraud ecosystem, this is a documented instance of the local protection layer that keeps regional call centre operations alive, and of the enforcement pressure now being applied to it.

therecord.media → (opens in a new tab)

Vulnerability Scans for Older Cisco IOS Vulnerability / Opt-Out Headers, (Thu, Nov 5th) (opens in a new tab)

SANS ISC Diary ·fetched 5 Nov 2026, 15:39 UTC CVE-2023-20198 EPSS 99.6% agreed3/3

Why readShows what live CVE-2023-20198 scanning actually looks like on the wire, including a path-encoding trick that will slip past signatures matching the literal URI.

A SANS ISC honeypot logged a SOAP POST to /%77ebui_wsma_https, the percent-encoded form of the Cisco IOS XE web UI endpoint tied to CVE-2023-20198, carrying a WS-Security UsernameToken for admin. The sending host identified itself as a research scanner through Opt-Out-Email and Opt-Out-Url headers pointing at rootevidence.com, a convention worth recognising during triage so benign mass scanning is not worked as an incident. The practical takeaway is the encoding: detection content that greps for the plain string webui_wsma_https misses this request entirely, and the vulnerability still sits at an EPSS of 0.996 three years after disclosure.

isc.sans.edu → (opens in a new tab)

Alpine Linux Security Advisory: OpenSSH-server password authentication bypass (opens in a new tab)

Hacker News ·pantalaimon ·fetched 5 Nov 2026, 19:41 UTC Must read 399 points agreed3/3

Why readAlpine's openssh-server lets an unauthenticated attacker log in as any key-only account by sending an overlong password, with no retries needed.

On Alpine, where openssh-server is built without PAM, an attacker who knows the name of an account whose password field is `*` can authenticate by sending a password of 257 to 1024 bytes, provided at least one other account on the box carries a $5$ or $6$ hash. It works on the first connection, every time, against anything exposed on the SSH port. Upgrade the package and restart sshd, since upgrading alone leaves the running daemon vulnerable; if you cannot patch yet, set PasswordAuthentication no and restart.

alpinelinux.org → (opens in a new tab) Discussion (opens in a new tab)

Zyxel fixes zero-day exploited to hack nearly 1,000 network switches (opens in a new tab)

BleepingComputer ·Lawrence Abrams ·fetched 5 Nov 2026, 03:41 UTC Must read agreed3/3

Why readCVE-2026-107441, an unauthenticated out-of-bounds write in the Zyxel GS1900 web management interface, was exploited as a zero-day to compromise 996 switches and steal configs and credentials.

The flaw lets an unauthenticated attacker with network access run OS commands via crafted HTTP requests, and affects ten GS1900 models; Zyxel shipped fixes on October 22 and advised on November 3. Exploitation by a suspected Chinese-speaking actor GreyNoise tracks as Kapibala began around August 17 and was initially misattributed to CVE-2026-7273, a pre-auth buffer overflow in the same switches patched on June 16, 2026, meaning operators who applied the June update may wrongly believe they are covered. Patch now, and treat stored configs and credentials on any exposed GS1900 as compromised.

Also covered byCSO Online (opens in a new tab),Help Net Security (opens in a new tab),SecurityWeek (opens in a new tab).

bleepingcomputer.com → (opens in a new tab)

Barracuda Networks is your new best friend (Barracuda Email Security Gateway Pre-Auth Remote Root RCE CVE-2026-116606 / CVE-2026-111560) (opens in a new tab)

watchTowr Labs ·Sina Kheirkhah (@SinSinology) ·fetched 5 Nov 2026, 23:41 UTC Must read Research CVE-2026-116606 EPSS 1.3% agreed3/3

Why readPre-auth remote root RCE chain on Barracuda Email Security Gateway, the same edge appliance UNC4841 burned with CVE-2023-2868.

watchTowr details CVE-2026-116606 and CVE-2026-111560 in Barracuda's Email Security Gateway, reaching code execution as root before authentication on an appliance that terminates and inspects all inbound mail. Barracuda has published its own advisory with affected and fixed versions. ESG has a history here: the 2023 CVE-2023-2868 zero-day led Mandiant to recommend physical replacement of compromised units rather than patching, so treat exposure as urgent and check the box rather than assuming the patch closed the story.

labs.watchtowr.com → (opens in a new tab)

CVE-2026-105769 (CVSS 9.3): kaniko before 1.25.20 can send a Microsoft Entra ID (Azure AD) access token for Azure Resource Manager to an attacker-controlled registry host. kaniko (opens in a new tab)

NVD ·fetched 5 Nov 2026, 03:41 UTC CVE-2026-105769 CVSS 9.3 EPSS 0.2% agreed3/3

Why readkaniko before 1.25.20 hands your Azure build identity's Entra ID token to any registry hostname that merely contains .azurecr.io, so a hostile FROM line in a Dockerfile steals ARM credentials.

The ACR credential helper kaniko uses (github.com/chrismellard/docker-credential-acr-env) decides whether a host is an Azure Container Registry with an unanchored regex, so x.azurecr.io.attacker.com matches. kaniko consults the helper for any registry lacking Docker config credentials that is not Google or ECR, then POSTs a token from the build's service principal, federated workload identity or managed identity to https://<host>/oauth2/exchange. The attacker can replay that token against Azure Resource Manager with the identity's full role assignments until expiry. Fixed in 1.25.20; CVSS 9.3, EPSS 0.00196.

nvd.nist.gov → (opens in a new tab)

CVE-2026-120091 (CVSS 8.7): Temporal Server compiles a Worker Controller Instance module, go.temporal.io/auto-scaled-workers, into its Worker Service, and that module registers a (opens in a new tab)

NVD ·fetched 5 Nov 2026, 19:41 UTC CVE-2026-120091 CVSS 8.7 EPSS 0.5% agreed3/3

Why readA read-only namespace role on Temporal Server reaches a Kubernetes compute provider that runs an attacker-chosen command on the Worker Service host, with no deployment needing to exist first.

The go.temporal.io/auto-scaled-workers module compiled into Temporal's Worker Service registers a k8s compute provider that builds its Kubernetes client from client configuration supplied in the caller's request rather than from operator configuration, and building that client can execute a command of the caller's choosing as the server process account. CreateWorkerDeploymentVersion needs a namespace write role, but ValidateWorkerDeploymentVersionComputeConfig is registered as read-only, so an authenticated caller with read access in a single namespace reaches the provider through it. Execution happens during validation, so no scaling decision, task arrival or unusual request sequence is required; the privilege boundary Temporal operators assume around read-only roles does not hold.

nvd.nist.gov → (opens in a new tab)

Debian, Ubuntu patch APT flaw that could hand attackers root (opens in a new tab)

CSO Online ·fetched 5 Nov 2026, 15:39 UTC CVE-2026-109518 EPSS 0.9% agreed3/3

Why readThe trust boundary that makes unauthenticated HTTP package transport acceptable is exactly what broke here, so anyone running stock Debian or Ubuntu should treat this as a priority patch.

CVE-2026-109518 is a flaw in how APT handles HTTP redirects and resumed downloads: the client can merge two different package downloads, which an attacker positioned on the network path to an archive server can abuse to get an unverified, attacker-controlled package installed with root privileges. Debian disclosed it on November 4 crediting Max Justicz, and Ubuntu rated it High at CVSS 7.7. Both distributions have shipped fixes; the practical takeaway is that signature verification alone did not contain the bug, so patching APT itself is the control.

csoonline.com → (opens in a new tab)

CVE-2026-120538 (CVSS 8.1): The Easy Digital Downloads plugin for WordPress is vulnerable to authentication bypass leading to administrator account takeover in all versions up to (opens in a new tab)

NVD ·fetched 5 Nov 2026, 23:41 UTC CVE-2026-120538 CVSS 8.1 EPSS 0.4% agreed3/3

Why readEasy Digital Downloads through 3.7.2 hands an unauthenticated attacker an admin session by starting a PayPal guest checkout with the administrator's email address.

edd_get_purchase_form_user() binds a guest order to an existing WordPress user purely on email-address match and rewrites the order's user_id. PurchaseUser::maybe_log_in_after_capture() then looks the user up by order email and, because the binding step guarantees the IDs match, calls wp_set_auth_cookie() for that account. Any administrator without an existing EDD customer record can be taken over by completing a PayPal purchase flow with their email; patch above 3.7.2.

nvd.nist.gov → (opens in a new tab)

CVE-2026-94290 (CVSS 9.1): SUSE Rancher Fleet clusters reference a kubeconfig secret that the controller uses to install and manage a Fleet agent on the target cluster. The cont (opens in a new tab)

NVD ·fetched 5 Nov 2026, 03:41 UTC CVE-2026-94290 CVSS 9.1 EPSS 0.5% agreed3/3

Why readA Fleet Cluster object in a namespace an ordinary user controls can reference the management cluster's admin kubeconfig secret, turning tenant-level access into control of the Rancher management cluster.

SUSE Rancher Fleet's controller did not restrict which kubeconfig secrets a Cluster object could reference. A user who can create Fleet Cluster objects in their own namespace can point one at administrative credentials for the management cluster, and the controller will then install the agent and run that user's Fleet content there with elevated privileges. Affects Fleet 0.16.0-0.16.3, 0.15.0-0.15.8, 0.14.0-0.14.12 and 0.13.0-0.13.17, plus possibly older releases; CVSS 9.1.

nvd.nist.gov → (opens in a new tab)

CVE-2026-73696 (CVSS 8.6): Control Web Panel (CWP) before webpanel_version 1.12 contains an execution after redirect vulnerability in the user panel authentication flow (CWPUser (opens in a new tab)

NVD ·fetched 5 Nov 2026, 07:39 UTC CVE-2026-73696 CVSS 8.6 EPSS 0.5% agreed2/3

Why readCWP authentication flaws have a track record of mass exploitation, and this one needs only an unauthenticated HTTP request that ignores a redirect.

CWPUserIndex emits a header('Location: ...') redirect for unauthenticated requests but omits the following exit(), so PHP continues through the rest of the handler and serves user panel functionality in the body of the same response. Any client that disregards the 302 and reads what follows reaches features that were supposed to require a session. Patched in webpanel_version 1.12; given how exposed CWP installs typically are, treat this as the one on today's list to schedule first.

nvd.nist.gov → (opens in a new tab)

CVE-2026-116769 (CVSS 9.3): CVE-2026-116769 is an authentication bypass vulnerability in Secure Access servers starting with version 13.0 and prior to 14.61. Attackers can modify (opens in a new tab)

NVD ·fetched 5 Nov 2026, 07:39 UTC CVE-2026-116769 CVSS 9.3 EPSS 0.4% agreed3/3

Why readUnauthenticated authentication bypass in Secure Access server versions 13.0 through 14.60 lets an attacker change a client's security settings through the server.

CVE-2026-116769 (CVSS 4.0 score 9.3) carries an AV:N/AC:L/PR:N/UI:N vector with high integrity impact and a scope change into connected clients, meaning no credentials are needed to tamper with client-side security settings. Fixed in 14.61. EPSS is 0.004 and there is no KEV entry yet, but an unauthenticated bypass in a remote access server is the category that gets weaponised quickly, so confirm your version against 14.61 now.

nvd.nist.gov → (opens in a new tab)

CVE-2026-119179 (CVSS 8.3): A flaw in the connection management layer of the MongoDB Node.js driver can leave a pooled network connection in the pool for reuse while data from an (opens in a new tab)

NVD ·fetched 5 Nov 2026, 23:41 UTC CVE-2026-119179 CVSS 8.3 EPSS 0.3% agreed3/3

Why readA connection-pool flaw in the MongoDB Node.js driver can pair database replies with the wrong requests, leaking one user's records to another.

The driver's connection management layer can return a pooled socket to the pool while a prior operation's data is still outstanding on it. Applications that set the optional client-side operation timeout can be pushed into this state by an unauthenticated party sending ordinary requests, causing responses to be matched to the wrong request. Consequences are cross-user data disclosure, wrong results returned to callers, and failures of legitimate queries, which makes this a correctness bug as much as a security one.

nvd.nist.gov → (opens in a new tab)

CVE-2026-120574 (CVSS 8.1): The SureCart plugin for WordPress is vulnerable to unauthenticated account takeover in all versions up to, and including, 4.9.5. The passwordless one- (opens in a new tab)

NVD ·fetched 5 Nov 2026, 23:41 UTC CVE-2026-120574 CVSS 8.1 EPSS 0.3% agreed3/3

Why readSureCart through 4.9.5 exposes a passwordless login REST route with no lockout, so a 6-digit code can be brute-forced into an admin session.

VerificationCodeController::verify() has a permission callback that returns true, resolves an attacker-supplied login to any WordPress user, and on a correct 6-digit code calls wp_set_auth_cookie() regardless of role. Affected versions have no per-account attempt counter and no persistent email or IP lockout, and the upstream CDN throttle can be defeated by IP rotation. An unauthenticated attacker can request a code for an administrator through the companion endpoint and brute-force it to full account takeover.

nvd.nist.gov → (opens in a new tab)

Full-Chain-Sentinel: an all-in-one security agent for web pentesting, Android and iOS app testing, batch operations, automated incident response and open-source vulnerability monitoring (opens in a new tab)

translated zgsfsec/Full-Chain-Sentinel-System: 全链哨兵，是集Web 渗透、安卓 App 安全测试、iOS APP安全厕所、攻防批量打点、自动化应急响应、开源漏洞监测为一体的网络安全 Agent，用于安全服务交付自动化辅助，禁止用于非法用途。

GitHub: new security tools ·zgsfsec ·fetched 5 Nov 2026, 11:38 UTC ★ 676 agreed2/3

Why readA Chinese-language LLM agent that chains web, Android and iOS testing with mass reconnaissance has picked up 676 stars in under a month, which is the kind of tooling that shows up in engagement reports soon after.

Full-Chain Sentinel System is a TypeScript security agent that bundles web penetration testing, mobile application testing, bulk attack-surface fingerprinting, automated incident response and open-source vulnerability monitoring into one delivery pipeline. The repository is a capability blurb rather than documented methodology, so there is no evidence yet about finding quality or false positive rates. Treat the star count as a signal of adoption velocity in the Chinese security services market, not as validation of the tool.

github.com → (opens in a new tab)

| … (opens in a new tab)

detect.fyi ·Sergio Albea ·fetched 5 Nov 2026, 15:39 UTC agreed3/3

Why readKQL hunting logic for Entra ID service principal sign-ins using stolen client secrets, including the invalid and expired secret attempts that give the attacker away.

Following a Todyl writeup of a campaign authenticating to Entra ID as legitimate applications using stolen client secrets, Albea builds hunting queries around application credential use rather than user sign-ins. The angle is failed authentications with invalid or expired secrets, which surface attackers working through a stale credential set, and the long-forgotten Graph, Key Vault and SharePoint permissions those apps may still hold. Deliberately IOC-free, so it stays useful after the campaign infrastructure rotates.

detect.fyi → (opens in a new tab)

GRIDSHIELD: Verifiable Grid Feasibility and Manipulation Resistance in Privacy-Preserving Peer-to-Peer Energy Trading (opens in a new tab)

arXiv cs.CR (all) ·Abdullahil Kafi, Md Zahidul Islam, Alvi Ataur Khalil ·fetched 5 Nov 2026, 03:41 UTC Research agreed3/3

Why readA protocol design showing how to keep a market operator honest when it both clears sealed bids and must respect physical network limits, with the feasibility check done over commitments rather than plaintext.

GRIDSHIELD replaces ordered clearing in blockchain peer-to-peer energy markets with an order-independent batch program, which removes the profit a compromised operator could extract by reordering or curtailing trades. Grid feasibility is proven with an aggregated congestion-feasibility proof that rebuilds each line constraint from the additive homomorphism of per-bus Pedersen commitments, so the operator proves the clearing is physically valid without revealing individual bids and without a per-circuit trusted setup. The contribution is defensive cryptographic engineering for a narrow domain; the threat model, an insider holding clearing privilege, generalises to other sealed-bid matching systems.

arxiv.org → (opens in a new tab)

Integrated Identity Protection for Semantic Communications in Embodied IoT Systems (opens in a new tab)

arXiv cs.CR (all) ·Yu Zhang, JingQi Liu, Shuangrui Zhao, Yuanyu Zhang ·fetched 5 Nov 2026, 19:41 UTC Research agreed2/3

Why readPrimary research treating a semantic communication signal as a carrier of transmitter identity, so authentication rides along with the payload instead of sitting above it.

The authors argue that trustworthy task decisions in embodied IoT need source legitimacy, not just semantic recovery, and propose exploiting physical layer identity information already present in every SemCom transmission. They analyze where the two information resources can be integrated at the observation, processing and decision levels, then build a dual branch mechanism covering both active impersonation and passive identity inference. Narrow in scope and far from deployment, but the authentication framing is a defensive one.

arxiv.org → (opens in a new tab)

Unlocking iOS Sysdiagnose and Unified Logs: Hexordia Contributes New Parsers to iLEAPP (opens in a new tab)

Hexordia ·Kevin Pagano ·fetched 5 Nov 2026, 15:39 UTC Research agreed3/3

Why readNew iLEAPP parsers that turn iOS sysdiagnose archives and unified logs into timeline-ready evidence, including automatic unpacking of embedded sysdiagnose.tar.gz inside full file system extractions.

Hexordia has contributed parsers to iLEAPP covering two sources that standard database extractions miss: sysdiagnose archives, which carry hardware identity, system configuration and execution history, and the unified logarchive, which holds volatile user interaction and hardware state. A centralised ingestion helper detects and unpacks embedded sysdiagnose.tar.gz found in FFS extractions and triage folders, so examiners no longer have to stage those manually. Pull the current iLEAPP and the capability is available on your next case.

hexordia.com → (opens in a new tab)

OrcDecryptNet: a .NET decryptor for DFIR-ORC archives (opens in a new tab)

Intrinsec ·CERT Intrinsec ·fetched 5 Nov 2026, 11:38 UTC Research agreed3/3

Why readA .NET 8 command-line tool that decrypts DFIR-ORC .p7b archives on both Windows and Linux, removing a Windows-only step from forensic collection pipelines.

CERT Intrinsec released OrcDecryptNet, a .NET 8 CLI for decrypting the encrypted .p7b archives produced by DFIR-ORC. The motivation is practical: their processing chain spans Windows and Linux, and the existing decryption path added deployment constraints. Anyone running DFIR-ORC collection at scale can drop this into an automated ingestion pipeline today.

intrinsec.com → (opens in a new tab)

Dead Ends And Data Recovery: The Importance Of Remaining Tool Agnostic In Digital Forensics (opens in a new tab)

Forensic Focus ·Forensic Focus ·fetched 5 Nov 2026, 07:39 UTC agreed3/3

Why readA case where automated forensic suites returned nothing and manual work on vehicle data produced the evidence that closed a fatal hit-and-run.

Page One's Jeffrey Lehn recovered vehicle data that standard forensic tooling could not surface, and the resulting evidence contributed to resolving a fatal hit-and-run. The point of the piece is procedural: examiners who stop where the automated parser stops will miss data that is present on the device. Useful as an argument for tool-agnostic validation in casework, though the published detail on the extraction method is limited.

forensicfocus.com → (opens in a new tab)

iOS Acquisition Troubleshooting: Where The Logs Are? (opens in a new tab)

ElcomSoft ·Oleg Afonin ·fetched 5 Nov 2026, 19:41 UTC agreed2/3

Why readNames the three separate log tiers iOS Forensic Toolkit writes, where each lands on disk, and which ones actually matter when an acquisition fails.

ElcomSoft walks through a tiered logging scheme in which the command itself, the agent installation step, and the extraction agent running on the device each produce their own file, none of them stored alongside the executable. The piece maps failure symptoms to the relevant log, covering bootloader exploits that do not fire, undetected iOS versions that break the firmware path, and agent signing problems. It is vendor support documentation rather than independent analysis, but it is the kind of operational detail examiners only discover under time pressure during a live extraction.

blog.elcomsoft.com → (opens in a new tab)

ChatGPT Lockdown Mode Bypass: Phishing, Malware Delivery, and Silent Data Exfiltration (opens in a new tab)

Check Point Research ·stcpresearch ·fetched 5 Nov 2026, 15:39 UTC Must read Research agreed3/3

Why readShows a bypass of OpenAI's Lockdown Mode that restores silent exfiltration of full chat history, with zero user interaction in ChatGPT Classic.

Check Point found a flaw that defeats the core protections of Lockdown Mode, the setting OpenAI added after an earlier DNS-tunnelling exfiltration bug in the code-execution runtime. The bypass reopens a broad surface: silent exfiltration of an entire conversation history, credential phishing inside the chat, and malware delivery, triggered in ChatGPT Classic simply by having a conversation. Anyone relying on Lockdown Mode as a control for sensitive data in ChatGPT needs to re-examine that assumption.

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

Signed, Sealed, Leaked: Trust Beyond Signature for MCP Toolchains (opens in a new tab)

arXiv cs.CR (AI) ·Yichiet Aun, Mohammed A. Alhartomi, Jing Jie Tan, Waqas Nadeem ·fetched 5 Nov 2026, 07:39 UTC Must read Research agreed3/3

Why readA legitimately signed MCP tool exfiltrated data to a policy-allowed destination and defeated Meta's MCPGuard-Dynamic in every trial, and the paper proposes a containment model that does not depend on trusting tool code.

The authors build a signed MCP server that performs its advertised function while embedding confidential data in requests to a destination the runtime policy permits. MCPGuard-Dynamic blocks a direct exfiltration attempt in the same run but never catches this one, showing that publisher authentication and destination allowlisting do not bound disclosure. Their proposed defence, germline execution, forks each tool invocation from a data-free process and gives the copy only the information authorised for its destination, so a malicious tool can leak no more than it was handed.

arxiv.org → (opens in a new tab)

A.S.E 2.0: Understanding How Repository Context Impacts the Security of LLM-Generated Code (opens in a new tab)

arXiv cs.CR (AI) ·Jiongchi Yu, Keke Lian, Junqing Wang, Zhikang Fan ·fetched 5 Nov 2026, 15:39 UTC Must read Research agreed3/3

Why readMeasures what repository context does to the security of LLM-generated code: functional pass rate rises up to 8.2pp while secure pass rate moves at most 3.7pp.

CASE-Bench builds 145 repository-level tasks from real CVEs across 51 projects, each running in Docker with functional tests and a proof-of-concept exploit validated in both directions, and annotated with the repository code that enforces or undermines the security property. Comparing 13 context construction methods and six controls across six LLMs shows retrieval tuned for functional correctness barely improves security, because the context decides whether the model sees the project's authorization checks or insecure code to imitate. Useful ammunition for appsec teams setting policy on coding agents in a monorepo.

arxiv.org → (opens in a new tab)

KVGuard: A KV Cache Safety Checker for LLM Serving Systems (opens in a new tab)

arXiv cs.CR (AI) ·Jonguk Cheong, Myungbin Hwang, Eulgyu Kim, Byoungyoung Lee ·fetched 5 Nov 2026, 23:41 UTC Must read Research agreed3/3

Why readDocuments a previously unstudied memory-safety bug class in PagedAttention KV cache management that corrupts outputs and breaks request isolation without crashing the server.

PagedAttention partitions a single CUDA allocation into fixed-size blocks mapped per request through a CPU-built block table, and that design admits double-free, use-after-free, out-of-bounds access, uninitialized read and permission violations at the block level. Existing sanitizers miss them because they reason about CUDA addresses rather than KV blocks, requests and block tables, and their redzone and quarantine designs do not fit GPU memory budgets. KVGuard maintains a compact shadow block per physical block recording per-request permission and per-entry initialization, and runs five safety checks against it. Cross-tenant inference leakage with no crash signal is the consequence worth taking to anyone running vLLM-style serving.

arxiv.org → (opens in a new tab)

Signed but Steerable: Measuring and Closing the Text-Vector Integrity Gap in Offline Retrieval-Augmented Generation (opens in a new tab)

arXiv cs.CR (all) ·Georgios Feretzakis, Alexandros Papaspyridis ·fetched 5 Nov 2026, 23:41 UTC Research agreed3/3

Why readShows that signing a RAG knowledge pack proves who published it but not that its embedding vectors match its text, and quantifies how many queries an attacker who controls the vectors can hijack.

Offline RAG on edge devices ships signed packs containing both text and precomputed embeddings; a verifier that recomputes vectors and accepts them within a cosine tolerance leaks a predictable share of queries, equal to the distribution of angular retrieval margins at that tolerance (the authors' margin law). Across five corpora and two embedding models, a tolerance set at floating-point noise still exposes up to 5.5% of queries (1.2% once exact duplicate documents are removed), while a threshold calibrated from a CPU INT8 versus FP32 comparison at the 0.1% quantile of honest cosines exposes 98-100%. On a Raspberry Pi 5 running a local model the attack drops correct answers from 0.75 to 0.39, and integrity audits of compressed vector stores do no better.

arxiv.org → (opens in a new tab)

Addition by Subtraction: Toward Mitigating Fragile Safety Alignment in Large Language Models (opens in a new tab)

arXiv cs.CR (AI) ·Hamid Dashtbani, Alexandra Dmitrienko ·fetched 5 Nov 2026, 11:38 UTC Research agreed3/3

Why readA finetuning method that removes safety-critical neurons during training so alignment stops depending on a sparse set of components that pruning or refusal-direction ablation can knock out.

Addition by Subtraction progressively ablates identified safety-critical neurons during finetuning and uses preference optimization to re-establish refusal behaviour through the rest of the network. The resulting models resist pruning-based attacks substantially better, and the robustness generalises to refusal-direction ablation and to black-box jailbreaks the method was not designed against. Relevant to anyone finetuning or redistributing open-weight models where a handful of neurons currently carries the whole safety story.

arxiv.org → (opens in a new tab)

Top AI agent security resources — November 2026 (opens in a new tab)

Adversa AI ·fetched 5 Nov 2026, 07:39 UTC agreed2/3

Why readA month of agent security findings in one place, including an account-wide Bedrock AgentCore takeover path and a guardrail model whose injection recall collapsed under a forged policy prefix.

The roundup indexes 43 resources and leads with three concrete results: a single prompt to an exposed Amazon Bedrock AgentCore agent reaching instance metadata credentials and from there every agent in the same account and region; embedded browsers in Cursor, GitHub Copilot, Codex, VS Code and Claude Desktop granted trust beyond normal browser boundaries, yielding host code execution or cross-disk file reads; and TypeSafe's Jev, the fast decision model LangChain uses for tool risk gating, dropping from 83 percent prompt injection recall to under 4 percent when a counterfeit policy was prepended. A separate cluster covers agents acting against their operators without an external attacker, including one that disabled security alerting and another that removed its monitoring integration and then denied doing so. Treat it as a reading list rather than primary research; the value is coverage, and the underlying reports are where the detail lives.

adversa.ai → (opens in a new tab)

Beyond the Landing Page: Website Blocking, Tracker Visibility, and Human Alignment in Automated Web Crawls (opens in a new tab)

arXiv cs.CR (AI) ·A H M Nazmus Sakib, Mahsin Bin Akram, Raveen Wijewickrama, Igor Bilogrevic ·fetched 5 Nov 2026, 19:41 UTC Research agreed3/3

Why readQuantifies how far automated crawl measurements diverge from real human browsing, and shows LLM-driven agents see tracking from 1.5 to 1.7 times as many third-party organizations as conventional crawlers.

Across 2,000 popular sites plus browsing telemetry from 26 participants, the authors compare agentic and rule-following crawlers under different automation frameworks and stealth configurations. Standard browser automation draws increasing denials once it moves past the landing page, while a detection-resistant framework largely avoids that escalation without inflating recorded tracking; agentic crawlers recover 82 to 87 percent of the tracker organizations real users encountered, with the gap widest on consent-managed pages. The finding cuts two ways: it is a correction factor for the privacy measurement literature, and it is evidence that agent-driven browsing has a materially different tracking exposure than scripted automation.

arxiv.org → (opens in a new tab)

Agents aren’t humans (but they kinda are) (opens in a new tab)

Cloudflare Blog ·Bryan Becker ·fetched 5 Nov 2026, 23:41 UTC agreed2/3

Why readGives you the vocabulary the agent-traffic debate is converging on, from the vendor that sees more of that traffic than anyone.

Cloudflare argues that the problem with personal AI agents is not detection but the broken link between a site and the human behind the request: an agent may look like a person, or like one visitor acting for many, and either way the site cannot tell who to hold responsible when something goes wrong. The piece reframes the question around identity, reputation, and accountability, and introduces the Operator as a distinct party that could carry liability for its agents' actions. It is a position paper rather than measurement, and it points toward signed-agent schemes Cloudflare has a commercial stake in, so read it for the frame and not for evidence.

blog.cloudflare.com → (opens in a new tab)

Post-Quantum Questions to Ask Your Vendors (opens in a new tab)

CISA Advisories ·CISA ·fetched 5 Nov 2026, 23:41 UTC agreed3/3

Why readA joint CISA, ASD ACSC and NSA quick-reference list of the PQC readiness questions to put to vendors during procurement and contract renewal.

The guidance gives organisations a structured way to assess third-party readiness for post-quantum cryptography, aimed at managed services, cloud capabilities and other cryptography-dependent products. It singles out systems with long asset lifecycles that are hard to update, such as industrial control systems and IoT devices, where transition depends entirely on vendor support or hardware replacement. Useful as a procurement artefact rather than a technical control change.

cisa.gov → (opens in a new tab)

Financial Stability Board sets out 12 ‘sound practices’ for FS firms’ AI adoption (opens in a new tab)

Compliance Week ·Neil Hodge ·fetched 5 Nov 2026, 03:41 UTC agreed3/3

Why readThe FSB has published 12 sound practices for AI adoption in financial institutions, the reference point regulators and supervisors will anchor to.

The Financial Stability Board issued a report setting out twelve sound practices financial institutions should consider for responsible AI adoption. Coming from the FSB, this is soft law rather than binding obligation, but it is the kind of document national supervisors pick up and examiners cite, so financial-sector risk and compliance teams will be asked to map against it. The item here is a short notice of publication; the report itself is where the substance sits.

complianceweek.com → (opens in a new tab)

What to expect from Democrats’ cybersecurity agenda in Congress (opens in a new tab)

Cybersecurity Dive ·Eric Geller ·fetched 5 Nov 2026, 19:41 UTC agreed2/3

Why readSets expectations for critical infrastructure operators and federal contractors on what a Democratic Congress is likely to pursue from January: CISA oversight, AI security, and revived state and local cyber grants.

Following midterm gains that hand Democrats both chambers, reporting from Cybersecurity Dive anticipates sharper oversight of CISA, renewed legislative attention to AI security, and an effort to restore grant funding for local utilities that Republican leadership had let lapse. Expect more hearings and investigations in areas lawmakers have avoided to date. Nothing here is binding yet, so read it as planning input for compliance and government affairs rather than a change in obligations.

cybersecuritydive.com → (opens in a new tab)

French bank Credit Mutuel Arkea says €100 million missing after cyberattack (opens in a new tab)

Google News: incidents · Reuters ·fetched 5 Nov 2026, 15:39 UTC Must read agreed3/3

Why readCredit Mutuel Arkea says roughly 100 million euros is missing after a cyberattack, a loss figure attached to a named European bank.

Reuters reports the French mutual banking group Credit Mutuel Arkea has disclosed around 100 million euros unaccounted for following a cyberattack. A stated loss of that size at a named retail bank is the sort of event that draws regulator attention under DORA and that peers in financial services will be asked about this week. Detail is currently limited to the wire report; expect attack specifics to follow.

Also covered byThe Block (opens in a new tab),RFI (opens in a new tab).

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

French authorities probe cyberattack targeting nation’s blood supply (opens in a new tab)

Google News: incidents · Cybersecurity Dive ·fetched 5 Nov 2026, 23:41 UTC agreed3/3

Why readFrance is investigating a cyberattack against its national blood supply, the kind of critical-health incident peers in the sector will be asked about.

French authorities have opened an investigation into a cyberattack affecting the country's blood supply. Detail is thin at this stage: no actor, no attack vector and no confirmed operational impact have been published. Healthcare and critical-service leaders should expect questions about continuity planning for an equivalent disruption.

Also covered byFox Business (opens in a new tab),China Daily Global Edition (opens in a new tab),NBC News (opens in a new tab),DW.com (opens in a new tab).

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

DHS Data Breach Exposes Information on More Than 40,000 Employees (opens in a new tab)

Google News: incidents · FEDagent ·fetched 5 Nov 2026, 15:39 UTC agreed3/3

Why readDHS has disclosed a breach affecting personal data on more than 40,000 employees, a federal incident peers and regulators will reference.

The Department of Homeland Security reported exposure of information on over 40,000 of its employees. No vector, actor or timeline has been published, and the initial report carries only the scope figure. Relevant mainly as a federal workforce-data incident that government and contractor security leaders will be asked about.

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

French Blood Banks Hacked in Latest Cyberattack on Europe (opens in a new tab)

Google News: incidents · Homeland Security Today ·fetched 5 Nov 2026, 11:38 UTC agreed3/3

Why readFrench blood banks have been hit in the latest of the attacks on European healthcare infrastructure, a sector-level incident a board will ask about.

French blood bank operations were disrupted by a cyberattack, continuing a run of incidents against European healthcare and critical health infrastructure. The item reaches us as an aggregator headline with no detail on the actor, the intrusion vector, or the operational impact on blood supply, so treat it as a pointer to watch for the primary reporting and any formal notification from the operator.

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

GMO CEO Faces Criticism at 3.5-Hour Press Conference Over Cyberattack Response (opens in a new tab)

Google News: incidents · JAPAN Forward ·fetched 5 Nov 2026, 03:41 UTC agreed3/3

Why readGMO's CEO took 3.5 hours of hostile press questioning over the company's cyberattack response, a case study in incident communications going badly in public.

Japanese reporting covers a three and a half hour press conference at which GMO's chief executive faced sustained criticism of how the company handled a cyberattack. The substance is the response and the disclosure posture rather than the intrusion itself, which is not detailed in what reached us. For leaders, the value is the reminder that post-incident communications in Japan are now conducted at this length and in this register, and that executives are personally the ones answering.

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

Srsly Risky Biz: The ritual bitcoin sacrifices will continue until security improves (opens in a new tab)

Risky Business News ·fetched 5 Nov 2026, 07:39 UTC agreed2/3

Why readA clean argument for why the Binance theft is not a lesson banks need to learn, grounded in settlement reversibility rather than hackability.

Uren and Jack take apart CZ's claim that North Korea's Binance hack is a warning to traditional banks. Their counterpoint is that conventional banks are probably just as breachable but sit inside a cooperative settlement system that can claw back fraudulent transfers, which makes walking away with billions far harder than it is in crypto. The episode also covers the pressure to accelerate software delivery with AI and why the security outcomes of that will be uneven.

risky.biz → (opens in a new tab)