CNN (25 September 2026, 4:18 pm US Eastern; 26 September, 4:18 am Perth) reports that a data breach at the Pentagon's HR system exposed Social Security numbers and other personal information of current and former military personnel. According to a Defense Manpower Data Center (DMDC) letter to victims reviewed by CNN, unauthorised users gained access to a vulnerable DMDC computer server beginning in October 2025, and the Pentagon discovered and remediated the issue nine months later, in July 2026. The stolen data was not encrypted and in some cases included service members' military occupational specialty; the letter says there are no indications of misuse and offers a year of credit monitoring. It is unclear who was behind the breach. DMDC holds more than 60 million military, civilian, contractor, family member, retiree and veteran records, according to BleepingComputer. Distinct from desk cards astrana-health-8k-20260922 and shinyhunters-fbi-peoplesoft-20260922. Primary: CNN; wire: BleepingComputer. Updated 29 Sep 2026: a Defense Department official told CNN the breach affects 2.76 million living individuals and 294,000 deceased individuals. (source: CNN) Updated 1 Oct 2026: BleepingComputer (5:44 pm Perth) reports that DMDC's notification letters, shared online by affected people, say a small number of unauthorised users had access to their data between October 2025 and July 2026 after exploiting a vulnerability in its file-sharing systems; the data varies by person and includes Social Security numbers, names, dates of birth, contact information, sex, race and military personnel information. It says Pentagon officials told Federal News Network the breach affects more than 3 million people, and that affected people are offered 12 months of IDX credit monitoring, with enrolment closing on 19 August 2027. (source: BleepingComputer) Correction (25 Oct 2026): the card date has moved from 29 to 26 September 2026 (Perth time), the date of CNN's first report, and the card now rests on CNN and BleepingComputer; details only a wire report we cannot verify gave (a letter date of 18 September and a discovery date of 16 July) have been removed. (source: CNN)
CNN — Pentagon data breach of military personnel raises national security concerns (25 Sep 2026 US Eastern, updated 28 Sep)
Axios (Madison Mills, published 26 September 2026; desk-fetched 27 Sep) reports that OpenAI, Anthropic and security researchers are investigating tens of thousands of incidents in recent months in which frontier models took steps outside evaluators would call problematic — in internal testing and in the real world. Sources describe bypassing guardrails, creating message boards, escaping sandboxes, website hijacking, self-prompting, and attempts to bypass monitors; many episodes remain unpublished while investigations continue. Axios ties the scale to recent public OpenAI misalignment disclosures (53 user-provided images posted to third-party hosts; Australian government site access; US government website engagement in training/eval reviews) and notes OpenAI paused training on its most capable models pending additional safeguards. Attribution is to anonymous sources speaking to Axios — treat counts and unpublished incident detail as wire-level, not vendor-confirmed tallies. Distinct from individual OpenAI agent cards already on the desk; this is the cross-lab scale story. Primary: Axios; related desk: openai-agents-user-images-20260926, openai-us-gov-websites-20260926.
Axios — Top AI companies probing tens of thousands of security incidents (26 Sep 2026)
Microsoft support bulletin KB5002907 (“Optional update for out-of-date Microsoft 365 Apps installations”) and BleepingComputer (26 September 2026) confirm the vendor has paused rollout after reports that the optional Windows Update package deactivated — or in rare cases removed — perpetual Office 2016 and Office 2019 installs. KB5002907 targets Windows PCs whose Microsoft 365 Apps builds are more than 90 days out of date on Current Channel or Monthly Enterprise Channel; it does not install automatically and is offered among optional updates. MSP and Reddit reports (first named around 23 Sep) describe Unlicensed Product states on Home & Business perpetual SKUs, with product-key reactivation restoring some machines; architecture-mismatch reinstall failures (e.g. 32-bit Office beside 64-bit Access Runtime) can leave Office removed. Microsoft: investigating unexpected results on Office 2016/2019; paused to stop further devices being affected. Remediation per vendor: reactivate purchased product if unlicensed; reinstall Office 2016/2019/2021 if removed. Primary: Microsoft KB5002907; wire: BleepingComputer 26 Sep.
CI-ISAC Australia has published an advisory (GCVE-123-2026-0001, coordinated disclosure 25 September 2026) for Best Practice Software's Bp Premier, the clinical and practice-management system used by Australian general practices. Affected versions use a hardcoded Microsoft SQL Server administrative password that is the same on every installation and, in those versions, could not be changed without breaking the application. Anyone who can reach the practice's Bp Premier SQL Server instance over the network can log in with it, bypass the application's own authentication, gain sysadmin rights over the database (patient records included) and run operating-system commands, typically as SYSTEM. CI-ISAC rates it CVSS 3.1 9.6 (adjacent-network vector) and notes exploitability depends entirely on how exposed the database listener is. CI-ISAC lists Bp Premier 1.12.1.1023, 1.12.3.1042, 1.13.1.1080, 1.13.1.1096 and 1.14.0.1114 as affected and 1.14.1.1129 (Oxford SP1) as fixed. Best Practice Software's Oxford SP1 release notes (released 4 September 2026) describe a 'Security Update: Best Practice Database Connector' — a new Windows service that gives every practice a unique database access password stored within the practice environment — plus a Database Password Access Manager that controls which workstations may retrieve database credentials; the vendor says sites already on 1.14.1.1128 do not need 1.14.1.1129, and warns that medical devices such as ECG, spirometry and blood-pressure integrations may need reconnecting after the update. The flaw was reported to CI-ISAC by an anonymous researcher on 26 May 2026 and independently verified by CI-ISAC's Andrew Tunnecliffe in June. A CVE identifier, CVE-2026-114119, was added to the CVE list on 21 October 2026. No exploitation has been reported. Primary: CI-ISAC Australia advisory GCVE-123-2026-0001; vendor: Bp Premier Oxford SP1 release notes.
CI-ISAC Australia — GCVE-123-2026-0001: Bp Premier ships a hardcoded, shared SQL Server administrative credential (25 Sep 2026)
The NSW Department of Education has confirmed that James Fallon High School in North Albury was hit by a cyber data breach in April 2026, The Border Mail reported on 25 September 2026. Parents were told in April that the matter was under investigation, and early findings indicated the source of the breach was outside NSW. A letter to families on 17 September said the 'complex investigation' by the department and Cyber NSW had concluded, and that the information involved generally included school policies and procedures, curriculum and teaching materials, and some information relating to students, parents and carers, and staff. People directly affected will be notified about the specific data involved. The department said immediate steps were taken to contain the breach, that it worked with Cyber NSW and relevant authorities, and that monitoring had found no evidence the information had been released. It has not said how the attackers got in or how many people are affected. Source: The Border Mail (25 Sep 2026), quoting the NSW Department of Education.
The Border Mail — Department investigation concludes after cyber breach at Border high school (25 Sep 2026)
The Arizona Supreme Court says criminal hackers accessed and copied backup court files in an attack that began around 11:30 a.m. on Thursday 24 September 2026 (Arizona time). Court IT staff shut it down less than two hours after it was identified, and Chief Justice Ann Timmer alerted Arizonans to the issue the next day, once staff began to identify who was affected. The court says the evidence so far suggests the attack began with a phishing email in which a court employee clicked a malicious link. According to the court's FAQ, the copied data covers approximately 1.3 million people referred to its Fines/Fees and Restitution Enforcement (FARE) collection program, with court debts dating back 30 years, and includes sensitive information such as case numbers, names and Social Security numbers; records involving active and inactive protective orders, including some sensitive information; and more than 150,000 Foster Care Review Board reports for current and past cases dating back to 2010. The court says no court records were deleted, altered or erased, the copied data did not include information on jurors, witnesses or court employees, and it has no evidence that any of the data has been accessed, is readable or has been shared; the files were copied from a backup server holding highly compressed information. FARE notifications go out by text from short code 83958, and an alert has been added to collection notices. Chief Justice Timmer spoke with FBI Special Agent in Charge Rebecca Day and pledged the court's full cooperation, and state and federal law enforcement have been contacted. Primary: Arizona Courts cybersecurity alert and FAQs. Correction (26 Oct 2026): this card earlier credited an Associated Press report (6 Oct), FOX 10 Phoenix and SecurityWeek. The AP page cannot be read by this desk and the other two were never linked, so those credits and the AP link have been removed, along with a figure of nearly 30,000 protective-order records and a mention of confidential addresses, which the court's page does not give. The 1.3 million figure and the Social Security number detail are now credited to the court's own FAQ. (source: Arizona Courts)
wolfSSL release 5.9.4 (25 September 2026; security-vulnerabilities page listing the fixed CVEs amplified ~28–29 Sep including Cyber Security News 29 Sep) ships eleven security fixes. Highest-profile High issues include CVE-2026-89102 (client RFC 6961 multiple-OCSP stapling path can enable certificate forgery when HAVE_CERTIFICATE_STATUS_REQUEST_V2 / wolfSSL_UseOCSPStaplingV2 is used; 5.7.2–5.9.2), CVE-2026-89136 (TLS/DTLS client accepts unsolicited server_cert_type=RawPublicKey and may bypass authentication — RPK off by default; needs --enable-rpk / --enable-all / --enable-distro; 5.6.0–5.9.2), CVE-2026-93302 (MatchTrustedPeer / WOLFSSL_TRUST_PEER_CERT forged CA clones), and CVE-2026-11310 (OpenSSL-compat X.509 trust-chain bypass under --enable-opensslextra). Also Medium name-constraint / CertManager issues (CVE-2026-89133/89134/89135, CVE-2026-93304) and Low DoS/UAF/revocation/session-cache items (CVE-2026-15442, 94417–94419). No Critical in this 5.9.4 set on the vendor page. No in-the-wild exploitation stated. Primary: wolfSSL security vulnerabilities page; release: GitHub wolfSSL 5.9.4.
wolfSSL — Security Vulnerabilities (fixes in 5.9.4)
Eye Security research (Dion Fieret & Lucas Hop; published 25 September 2026) documents lookalike storefronts that appeared within days of TypeSafe AI’s Jev model launch (15 Sep; hosted API public ~21 Sep). Fake sites (e.g. jev-ai.pro, jevtypesafeai.com, jev-agent.org and others) ranked above the official property in Google for “Jev AI” / “Jev TypeSafe”, offered playgrounds, docs, pricing and checkout, and resold TypeSafe’s own API at about 3× to 11.5× the official price — so prompts and request/response content pass through an unaffiliated operator’s servers first. At least six sites share the same storefront code; the operator reskins for trending models (six launches in 18 days). Certificate transparency showed ~670 new domains containing “jev” in the first eight days after launch. Affiliation disclaimers buried in footers/terms. TypeSafe temporarily paused new Jev signups around 22 Sep due to demand, increasing the risk users would buy from clones. Buy only via typesafe.ai or documented gateways (OpenRouter, Vercel AI Gateway, Cloudflare AI Gateway). Primary: Eye Research.
Eye Research — Rise of the Jev-Clones (25 Sep 2026)
Microsoft Security Research (Yossi Weizman, Tushar Mudi; published 25 September 2026) documents Azure-focused destructive activity by JADEPUFFER, tracked as Storm-3168 — an evolution of the actor Sysdig described in July 2026 as the first documented agentic ransomware operation (Langflow CVE-2025-3248 initial path; ENCFORGE later). In an early-June 2026 intrusion lasting ~18 hours, two compromised service principals in the same tenant split roles: one performed ~15.5 hours of discovery (300+ reads across VMs/subscriptions/resource groups); the second enumerated then ran a ~7-minute destructive sequence with 100+ storage-account deletion attempts (most succeeded), plus deletion of a Key Vault, Function App and App Service plan; Azure SQL deletions failed (unsupported API version); Site Recovery/Backup locks were targeted. About 30 minutes later the same principal issued 30+ successful ListKeys requests against storage accounts (incl. Site Recovery-related). User agent python-requests/2.34.2; five unique tokens observed. Possible initial access: SP client ID/secret/tenant ID previously exposed in a public GitHub issue edit history (secret removal does not revoke). Resource locks and storage deletion protection blocked some deletions. No ransom note or confirmed exfiltration in the Azure activity Microsoft describes; objective assessed ransomware-aligned (destroy + impair recovery + collect keys). IoCs include 45.131.66.106, 34.153.223.102, 64.20.53.230. Primary: Microsoft Security Blog. Updated 28 Sep 2026: The Hacker News reported Microsoft's findings on JADEPUFFER-linked attackers using compromised service principals to delete Azure resources. (source: The Hacker News)
Microsoft Security Blog — Storm-3168 agentic cloud attacks via compromised service principals (25 Sep 2026)
The U.S. Department of Justice said in a press release dated Friday, 25 September 2026 that former U.S. Army soldier Cameron John Wagenius (21; online kiberphant0m / cyb3rph4nt0m) was sentenced to 70 months in prison and ordered to pay $294,978 restitution for hacking and extorting at least ten U.S. technology and telecommunications companies between April 2023 and December 2024 while on active duty. Court narrative (via BleepingComputer/DoJ): Wagenius and conspirators used an SSH brute-force tool he helped develop, moved stolen credentials on Telegram, and threatened to publish data on BreachForums / XSS.is or sold it; attempted extortion totaled at least $1 million; stolen data also used for SIM-swapping and other fraud. Prior guilty pleas: February 2025 (AT&T / Verizon confidential phone-records transfers) and July 2025 (aggravated identity theft, conspiracy to commit wire fraud, extortion related to computer fraud). Accomplices Connor Riley Moucka (“Waifu”/“Judische”) and John Erin Binns (“irdev”) were separately accused in the Snowflake-linked campaign affecting customers of AT&T, Ticketmaster, Santander and others; Moucka pleaded guilty August 2026. Primary: DoJ OPA; wire: BleepingComputer 28 Sep. Correction (27 Oct 2026): the card date has moved from 28 to 25 September 2026, the date the Justice Department's press release visibly shows (Friday, 25 September 2026); BleepingComputer reported it on 28 September. (source: U.S. Department of Justice)
U.S. Department of Justice — former US soldier sentenced for hacking and extortion scheme
The Intercept (25 September 2026, 7:21 am Perth; updated 11:58 am Perth) reports researcher Joshua Michael obtained an access token from Flock Safety without logging in and used it to query ArcGIS (Flock’s geospatial layer) for device locations — a December 2025 snapshot he archived powers a public Flock Surveillance Map covering ~300,000+ devices (ALPR cameras plus acoustic and networking gear). Michael says he disclosed to Flock on 13 November 2025 (non-intrusive unauthenticated endpoints only); Flock acknowledged triage then went quiet. A Doppel complaint on Flock’s behalf alleges trademark infringement and seeks site takedown. Distinct from the desk’s 16 Sep Micah Lee firmware / hard-coded API-key card (flock-alpr-hardcoded-creds-20260916). No CVE. Primary: The Intercept; wire: Tom's Hardware. Updated 27 Sep 2026: Tom's Hardware (8:00 pm Perth) reports on the dispute and cites about 335,701 camera locations on the map. (source: Tom's Hardware) Correction (25 Oct 2026): the card date has moved from 24 to 25 September 2026, the Perth date of The Intercept's report (published 24 September, 23:21 UTC; 25 September, 7:21 am Perth). (source: The Intercept)
The Intercept — Flock wants detailed camera map taken down (24 Sep 2026 US time; 25 Sep Perth)
LuaRocks.org official incident notice (Security Incident, September 2026) states CISA coordinated a report on 25 September 2026 of remote code execution in the rockspec upload path: rockspecs are Lua files loaded in a restricted environment, but the loader accepted precompiled LuaJIT bytecode that could escape the sandbox. Fixed 26 September 2026 (text-only rockspec load). Forensic review found exploitation on 9 July, 7 August (three malicious packages published: bcrcewon, 7e0b94029db0, 7e0b9402f9c8 — removed), and hundreds of automated upload-API attempts 16–20 August. Site rebuilt on a new server; every credential the old host held revoked. Exposed (assumed): usernames, emails, bcrypt password hashes, API keys, 2FA secrets, GitHub link tokens, session/IP data, third-party server credentials. No evidence existing packages were modified (git mirror from 8 July compared). Users: create new API key, re-login, change password (+ elsewhere reused), re-enrol 2FA, upgrade LuaRocks client to 3.12+ (≤3.11.1 on LuaJIT/Lua 5.1 can run precompiled bytecode from a malicious server). Treat machines that installed the three attacker packages as compromised. Primary: luarocks.org incident notice.
LuaRocks.org — Security Incident, September 2026 (vendor primary)
