Reporting Period: 26th September 2026 – 2nd October 2026
Arcanum’s View
This week’s activity clusters around three connected shifts. First, the internet-facing perimeter took sustained fire: Cisco, F5, Check Point and WordPress core all had flaws under active exploitation within days of disclosure or patch release [1]. Second, artificial intelligence moved from an attacker’s assistant to an attacker’s operator: a single actor ran an agentic breach campaign against 27 organisations for under $20,000 in compute costs [2], while the National Cyber Security Centre’s own weekly notes recorded autonomous AI agents exploiting access-control flaws in test environments [3]. Third, the money moved: a record month for ransomware [4] was matched by one of 2026’s largest single cryptocurrency exchange heists [5] and a data breach exposing over 23 million user records at a widely used screenshot platform [6]. None of this is a single dramatic event: it is a baseline shift, and it is the pattern, not any individual incident, that should worry security leaders most.
Immediate Priorities for Security Leaders
- Patch immediately: Cisco ISE, F5 BIG-IP APM, Check Point Quantum Gateway and Security Management, and WordPress core all have confirmed in-the-wild exploitation this week, alongside a fresh batch of CISA Known Exploited Vulnerabilities catalogue additions.
- Review the default data-handling settings of AI coding and agent tools in use across engineering, since two mainstream tools were found sending private repositories to vendor cloud infrastructure without clear consent.
- Reassess exposure to third-party vendors that store customer photos, documents or other PII, and confirm DNS and CDN change-control is locked down after this week’s high-profile domain hijack.
- Stress-test detection and containment for payment-rail and crypto hot-wallet compromise, and confirm whether reserve or insurance funds could absorb a comparable loss.
- Treat the gap between vulnerability disclosure and mass exploitation as hours, not days, when setting patching SLAs for internet-facing systems.
Detailed, ranked actions with owners and deadlines appear in the Priority Actions table at the end of this advisory.
Key Threats
Internet-facing infrastructure under coordinated exploitation
Cisco disclosed and patched a maximum-severity authentication bypass in Identity Services Engine (CVE-2026-76460, CVSS 9.5) after confirming unauthenticated attackers were using it to reach the management API and, from there, root access [7]. F5 confirmed active exploitation of a heap-based buffer overflow in BIG-IP’s Access Policy Manager (CVE-2026-94127, CVSS 9.3) affecting systems configured as OAuth authorisation servers, and the vendor has not disclosed how many systems were compromised [8]. Check Point disclosed two actively exploited flaws in Quantum Security Gateway and Security Management: a pre-authentication remote-code-execution bug via VPN certificate handling, under attack since 12 September against Spark firewall customers globally, and a path-traversal flaw enabling arbitrary script execution [9]. WordPress core’s CVE-2026-87902, a critical unauthenticated remote-code-execution flaw, was exploited within hours of the fix shipping on 22 September, with 68 exploitation attempts logged in the first day [10]. CISA also added a cluster of Linux kernel, Zyxel switch, Acronis backup-plugin and Google Pixel modem flaws to its Known Exploited Vulnerabilities catalogue this week, each carrying a federal remediation deadline [11]. Separately, SolarWinds pre-emptively patched a hard-coded cryptographic key in Access Rights Manager (CVE-2026-28326, CVSS 8.8) that would allow unauthenticated remote code execution, though no exploitation has yet been confirmed [12].
AI-enabled threat activity
Researchers documented a single operator running the open-source Hermes, Strix and Cairn agent frameworks, powered by commercial and open models, to autonomously scan, exploit and monetise 27 organisations, deploying payment-card skimmers on 19 websites and stealing over 600,000 card records for a total operating cost of $12,000 to $18,000 [13]. A separate campaign is flooding the web with fake support pages and optimised content designed to poison ChatGPT, Gemini and Google AI Overviews into surfacing fraudulent contact details for major airlines and banks, including Delta, Lufthansa, Chase and Bank of America [14]. AI coding assistants from two major vendors, including Z.ai’s ZCode tool, were found to send users’ local code repositories to the vendor’s own cloud infrastructure by default without clear consent, prompting one vendor to disable the workflow and open its code for public review [15]. A separate disclosure showed a Visual Studio Code flaw allowing a single click on an innocuous-looking link to bypass Workspace Trust and execute code with full user permissions, no exploit chain required [16]. The National Cyber Security Centre’s weekly threat notes this week also flagged autonomous AI agents identifying and exploiting access-control flaws in lab environments, and Chinese hacking groups using AI to scale espionage operations [17].
Large-scale data exposure
Screenshot and screen-recording platform Gyazo confirmed that attackers exploited a server vulnerability to steal 23.62 million user records, including password hashes, session identifiers and billing information, along with 490 million image-metadata records containing upload IP addresses, EXIF location data and OCR-extracted text; the company said it could not rule out that private images were viewed [18]. Publisher Elsevier had three core domains, including elsevier.com, briefly hijacked and redirected to a page branded “LAPSUS$ Group, Chapter II” for around 78 minutes, in what researchers believe was a DNS or CDN configuration change rather than a breach of core systems; Elsevier stated no customer data or research content was compromised [19].
Ransomware and extortion
NCC Group’s latest threat intelligence reporting shows ransomware hit a record 1,073 organisations in August 2026, a 12% increase on July and the second consecutive month at the highest levels recorded this year, with the industrial sector accounting for nearly a third of victims and the Qilin and “The Gentlemen” groups the most prolific attackers [20]. The Elsevier hijack’s LAPSUS$ “Chapter II” branding illustrates a related trend: researchers found no evidence linking the branding to the original 2021-2022 LAPSUS$ membership, suggesting either a genuine revival or an opportunistic use of a feared name to maximise psychological impact [21].
Financial and crypto infrastructure
Crypto exchange Bitget disclosed that attackers compromised a critical backend system and used it to spoof transaction data, moving approximately $351.6 million in Ether, XRP, BNB, AVAX, USDT and USDC out of hot and warm wallets before the intrusion was detected and contained; cold wallets were unaffected [22]. Bitget’s CEO said preliminary analysis points to North Korean involvement, and the exchange said customer losses would be covered by its User Protection Fund, which holds over $464 million [23].
Strategic Assessment
AI has compressed the attacker’s timeline from weeks to hours
This week’s evidence suggests the constraint on attackers is shifting from skill to orchestration. A 27-organisation breach campaign that would once have required a skilled team now runs on off-the-shelf agent frameworks for the cost of a modest cloud bill [24]. For defenders, the practical implication is that the gap between disclosure and mass exploitation can no longer be measured in days: WordPress’s newest critical flaw was under attack the same day the patch shipped [25], and patching SLAs built around a “weekly cycle” are already obsolete.
The perimeter, not the endpoint, is still the softest target
Every major infrastructure compromise this week, Cisco ISE, F5 BIG-IP APM and Check Point Quantum, sits at the network edge or in the identity path rather than on a user device [26]. These are exactly the systems organisations rely on to authenticate and segment everything else, so a single unpatched appliance can undo years of endpoint and identity investment elsewhere.
Trust infrastructure is becoming as valuable a target as data itself
The Elsevier domain hijack stole no data but still generated global coverage, showing that briefly controlling a trusted domain, or reviving a feared brand name, can be as disruptive to an organisation as an actual breach [27]. Boards should treat DNS, CDN and registrar access as a crown-jewel control, not a routine IT administration task.
Compensating controls, not just prevention, decided this week’s biggest financial loss
Bitget’s $351.6 million loss was contained to hot and warm wallets and absorbed entirely by a pre-funded reserve, meaning customers were made whole despite one of the year’s largest exchange breaches [28]. The lesson for any organisation handling high-value transactions is that detection speed and pre-funded recovery mechanisms can matter as much as prevention.
Assurance Considerations, Tied to This Week’s Findings
| Control area (CAF / GovAssure / NIS2) | This week’s evidence it matters |
|---|---|
| Vulnerability management & rapid remediation | Cisco ISE, F5 BIG-IP APM, Check Point Quantum and WordPress core were all exploited in the wild within days, or hours, of disclosure this week [29]. |
| Identity and access control | Cisco ISE’s authentication bypass and F5 BIG-IP APM’s role as an OAuth authorisation server both let unauthenticated attackers reach privileged functions [30]. |
| Secure-by-design AI deployment | AI coding assistants shipped with defaults that exfiltrated private repositories to vendor cloud infrastructure, and a Visual Studio Code flaw let a single link execute code with full user permissions [31]. |
| Supply-chain assurance | An AI agent campaign compromised 27 organisations end-to-end for under $20,000, and Elsevier’s domains were hijacked through what researchers believe was a DNS or CDN configuration change [32]. |
| Third-party/vendor governance | The Gyazo breach exposed 23.62 million user records held by a third-party vendor, underscoring the need to assess suppliers that store customer images and metadata [33]. |
| Detection and response | Bitget’s own monitoring detected and contained the intrusion to hot and warm wallet layers before attackers could reach cold storage [34]. |
| Operational resilience and recovery | Bitget’s pre-funded User Protection Fund absorbed the entire $351.6 million loss without customer impact, illustrating the value of funded recovery mechanisms over prevention alone [35]. |
Questions for Boards and Senior Leaders
- If our Cisco ISE, F5, or Check Point edge appliances were among those targeted this week, would we find out from our own monitoring or from a leak site?
- Do we have a documented, tested plan to detect and contain a hot-wallet or payment-rail compromise the way Bitget did, and would it actually hold up under a live attack?
- Which AI coding or agent tools has engineering adopted, and do we know their default data-handling settings for source code, credentials and customer data?
- If a vendor holding our customers’ images, documents or personal data were breached tomorrow, could we say which of our data was in scope within 24 hours, as Gyazo’s customers now need to?
- Could our own domains or CDN configuration be hijacked and rebranded, as happened to Elsevier, and would we detect it before 78 minutes had passed?
- Given a record month for ransomware and rising industrial-sector targeting, when did we last test our own backup isolation and recovery time against a realistic scenario?
Priority Actions, Ranked
| Severity | Finding | Action | Deadline |
|---|---|---|---|
| Critical | Cisco ISE, F5 BIG-IP APM, Check Point Quantum and WordPress core actively exploited | Patch or upgrade immediately; restrict management interfaces from the internet pending confirmation of patching | Within 48 hours |
| Critical | New CISA KEV additions (Linux kernel, Zyxel, Acronis, Google Pixel, SolarWinds ARM) | Confirm applicability, apply vendor patches, and validate against the federal remediation deadlines as a benchmark | Within 1 week |
| High | Backend compromise and transaction spoofing behind the Bitget heist | Review payment-rail and hot-wallet monitoring, transaction-signing integrity checks and reserve-fund coverage for equivalent exposure | Within 1 week |
| High | AI coding and agent tools exfiltrating repositories or bypassing workspace trust by default | Audit AI developer tool configurations across engineering; disable unreviewed cloud-sync defaults; patch or restrict affected IDE extensions | Within 2 weeks |
| Medium | Autonomous AI agent frameworks (Hermes, Strix, Cairn) exploiting common web flaws at scale | Prioritise remediation of SQL injection, sudo misconfiguration and exposed cloud credentials; assume automated, sub-24-hour exploitation timelines | Within 2 weeks |
| Medium | Third-party vendor exposure illustrated by the Gyazo breach | Inventory vendors holding customer PII, images or documents, and confirm breach-notification SLAs are contractually defined | Within 1 month |
| Medium | DNS/CDN hijacking illustrated by the Elsevier incident | Review change-control and privileged access for DNS, CDN and domain-registrar accounts | Within 1 month |
| Low | Record ransomware activity with rising industrial-sector targeting | Re-confirm backup isolation, immutability and recovery-time testing cadence at the next resilience review | Next board cycle |
[1]BleepingComputer — “Cisco warns of Identity Service Engine zero-day exploited in attacks” (17 September 2026) · The Register — “Someone’s attacking a critical 0-day RCE in F5 BIG-IP APM” (23 September 2026) · Truesec — “CVE-2026-85102 & CVE-2026-93616: Active Exploitation of Check Point Security Gateway and Security Management Vulnerabilities” (25 September 2026) · The Hacker News — “Attackers Exploit WordPress CVE-2026-87902 Within Hours of Disclosure” (22 September 2026) — https://www.bleepingcomputer.com/news/security/cisco-warns-of-identity-service-engine-zero-day-exploited-in-attacks/ · https://www.theregister.com/security/2026/09/23/someones-attacking-a-critical-0-day-rce-in-f5-big-ip-apm/5298659 · https://www.truesec.com/hub/blog/check-point-security-gateway-security-management-cve-2026-85102-cve-2026-93616 · https://thehackernews.com/2026/09/attackers-exploit-wordpress-cve-2026.html
[2]Tech Insider — “AI Agents Breach 27 Firms, Steal 600K Cards” (25 September 2026) — https://tech-insider.org/ai-agents-hermes-strix-cairn-breach-27-firms-2026/
[3]CTO at NCSC (Substack) — “CTO at NCSC Summary: week ending September 20th” (19 September 2026) — https://ctoatncsc.substack.com/p/cto-at-ncsc-summary-week-ending-september-06f
[4]Infosecurity Magazine — “Ransomware Attacks Reach Record High for 2026” (23 September 2026) — https://www.infosecurity-magazine.com/news/ransomware-attacks-reach-record/
[5]Insurance Journal — “Crypto Exchange Bitget Says Hackers Stole $352 Million” (25 September 2026) — https://www.insurancejournal.com/news/international/2026/09/25/886838.htm
[6]BleepingComputer — “Gyazo server flaw exploited to steal 23.6 million user records” (18 September 2026) — https://www.bleepingcomputer.com/news/security/gyazo-server-flaw-exploited-to-steal-236-million-user-records/
[7]Help Net Security — “Unauthenticated attackers are bypassing Cisco ISE’s management interface (CVE-2026-76460)” (17 September 2026) · BleepingComputer — “Cisco warns of Identity Service Engine zero-day exploited in attacks” (17 September 2026) — https://www.helpnetsecurity.com/2026/09/17/cisco-ise-vulnerability-exploited-cve-2026-76460/ · https://www.bleepingcomputer.com/news/security/cisco-warns-of-identity-service-engine-zero-day-exploited-in-attacks/
[8]The Register — “Someone’s attacking a critical 0-day RCE in F5 BIG-IP APM” (23 September 2026) — https://www.theregister.com/security/2026/09/23/someones-attacking-a-critical-0-day-rce-in-f5-big-ip-apm/5298659
[9]Truesec — “CVE-2026-85102 & CVE-2026-93616: Active Exploitation of Check Point Security Gateway and Security Management Vulnerabilities” (25 September 2026) — https://www.truesec.com/hub/blog/check-point-security-gateway-security-management-cve-2026-85102-cve-2026-93616
[10]The Hacker News — “Attackers Exploit WordPress CVE-2026-87902 Within Hours of Disclosure” (22 September 2026) — https://thehackernews.com/2026/09/attackers-exploit-wordpress-cve-2026.html
[11]CVE Brief — “CVE Brief – September 22, 2026” (22 September 2026) — https://cvebrief.com/archive/2026/09/22/
[12]The Hacker News — “SolarWinds Patches ARM Hard-Coded Key Flaw Enabling Unauthenticated RCE” (19 September 2026) — https://thehackernews.com/2026/09/solarwinds-patches-arm-hard-coded-key.html
[13]Tech Insider — “AI Agents Breach 27 Firms, Steal 600K Cards” (25 September 2026) — https://tech-insider.org/ai-agents-hermes-strix-cairn-breach-27-firms-2026/
[14]The Hacker News — “ThreatsDay: AI Search Poisoning, AI Coding Tool Leaking Repos, One-Click Code Execution and 13 More Stories” (24 September 2026) — https://thehackernews.com/2026/09/threatsday-ai-search-poisoning-ai.html
[15]The Hacker News — “ThreatsDay: AI Search Poisoning, AI Coding Tool Leaking Repos, One-Click Code Execution and 13 More Stories” (24 September 2026) — https://thehackernews.com/2026/09/threatsday-ai-search-poisoning-ai.html
[16]The Hacker News — “ThreatsDay: AI Search Poisoning, AI Coding Tool Leaking Repos, One-Click Code Execution and 13 More Stories” (24 September 2026) — https://thehackernews.com/2026/09/threatsday-ai-search-poisoning-ai.html
[17]CTO at NCSC (Substack) — “CTO at NCSC Summary: week ending September 20th” (19 September 2026) — https://ctoatncsc.substack.com/p/cto-at-ncsc-summary-week-ending-september-06f
[18]The Hacker News — “Gyazo Breach Exposes 23.62 Million User Records and 490 Million Image Metadata Records” (18 September 2026) · BleepingComputer — “Gyazo server flaw exploited to steal 23.6 million user records” (18 September 2026) — https://thehackernews.com/2026/09/gyazo-breach-exposes-2362-million-user.html · https://www.bleepingcomputer.com/news/security/gyazo-server-flaw-exploited-to-steal-236-million-user-records/
[19]Help Net Security — “Brief hijack makes Elsevier domains redirect to LAPSUS$ ‘Chapter II’ page” (22 September 2026) — https://www.helpnetsecurity.com/2026/09/22/elsevier-domains-hijack-lapsus/
[20]Infosecurity Magazine — “Ransomware Attacks Reach Record High for 2026” (23 September 2026) — https://www.infosecurity-magazine.com/news/ransomware-attacks-reach-record/
[21]Help Net Security — “Brief hijack makes Elsevier domains redirect to LAPSUS$ ‘Chapter II’ page” (22 September 2026) — https://www.helpnetsecurity.com/2026/09/22/elsevier-domains-hijack-lapsus/
[22]Insurance Journal — “Crypto Exchange Bitget Says Hackers Stole $352 Million” (25 September 2026) · BleepingComputer — “Hackers steal $351.6 million in Bitget crypto exchange hack” (25 September 2026) — https://www.insurancejournal.com/news/international/2026/09/25/886838.htm · https://www.bleepingcomputer.com/news/security/hackers-steal-3516-million-in-bitget-crypto-exchange-hack/
[23]Insurance Journal — “Crypto Exchange Bitget Says Hackers Stole $352 Million” (25 September 2026) — https://www.insurancejournal.com/news/international/2026/09/25/886838.htm
[24]Tech Insider — “AI Agents Breach 27 Firms, Steal 600K Cards” (25 September 2026) — https://tech-insider.org/ai-agents-hermes-strix-cairn-breach-27-firms-2026/
[25]The Hacker News — “Attackers Exploit WordPress CVE-2026-87902 Within Hours of Disclosure” (22 September 2026) — https://thehackernews.com/2026/09/attackers-exploit-wordpress-cve-2026.html
[26]Truesec — “CVE-2026-85102 & CVE-2026-93616: Active Exploitation of Check Point Security Gateway and Security Management Vulnerabilities” (25 September 2026) — https://www.truesec.com/hub/blog/check-point-security-gateway-security-management-cve-2026-85102-cve-2026-93616
[27]Help Net Security — “Brief hijack makes Elsevier domains redirect to LAPSUS$ ‘Chapter II’ page” (22 September 2026) — https://www.helpnetsecurity.com/2026/09/22/elsevier-domains-hijack-lapsus/
[28]Insurance Journal — “Crypto Exchange Bitget Says Hackers Stole $352 Million” (25 September 2026) — https://www.insurancejournal.com/news/international/2026/09/25/886838.htm
[29]The Hacker News — “Attackers Exploit WordPress CVE-2026-87902 Within Hours of Disclosure” (22 September 2026) · CVE Brief — “CVE Brief – September 22, 2026” (22 September 2026) — https://thehackernews.com/2026/09/attackers-exploit-wordpress-cve-2026.html · https://cvebrief.com/archive/2026/09/22/
[30]Help Net Security — “Unauthenticated attackers are bypassing Cisco ISE’s management interface (CVE-2026-76460)” (17 September 2026) · The Register — “Someone’s attacking a critical 0-day RCE in F5 BIG-IP APM” (23 September 2026) — https://www.helpnetsecurity.com/2026/09/17/cisco-ise-vulnerability-exploited-cve-2026-76460/ · https://www.theregister.com/security/2026/09/23/someones-attacking-a-critical-0-day-rce-in-f5-big-ip-apm/5298659
[31]The Hacker News — “ThreatsDay: AI Search Poisoning, AI Coding Tool Leaking Repos, One-Click Code Execution and 13 More Stories” (24 September 2026) — https://thehackernews.com/2026/09/threatsday-ai-search-poisoning-ai.html
[32]Tech Insider — “AI Agents Breach 27 Firms, Steal 600K Cards” (25 September 2026) · Help Net Security — “Brief hijack makes Elsevier domains redirect to LAPSUS$ ‘Chapter II’ page” (22 September 2026) — https://tech-insider.org/ai-agents-hermes-strix-cairn-breach-27-firms-2026/ · https://www.helpnetsecurity.com/2026/09/22/elsevier-domains-hijack-lapsus/
[33]BleepingComputer — “Gyazo server flaw exploited to steal 23.6 million user records” (18 September 2026) — https://www.bleepingcomputer.com/news/security/gyazo-server-flaw-exploited-to-steal-236-million-user-records/
[34]Insurance Journal — “Crypto Exchange Bitget Says Hackers Stole $352 Million” (25 September 2026) — https://www.insurancejournal.com/news/international/2026/09/25/886838.htm
[35]Insurance Journal — “Crypto Exchange Bitget Says Hackers Stole $352 Million” (25 September 2026) — https://www.insurancejournal.com/news/international/2026/09/25/886838.htm