ASD’s ACSC publication Agentic AI Harnesses — The layer above the model (first published 11 September 2026) defines the harness as every component of an agentic system other than the LLM itself (connectors, tool registry, memory, permissions). It complements the May 2026 Five Eyes careful-adoption guidance already on this desk. Core claim: some risks, including prompt injection, cannot be reliably mitigated inside the model alone — ASD states no fully reliable technical mitigation currently exists — so controls belong in the harness and connected systems (least privilege, human approval for high-impact actions, output verification, logging of prompts/tool calls/config changes). Audience: executives, CISOs, IT leaders adopting agentic AI. Primary: ACSC page + PDF. Amplify: iTnews (21 September 2026 AEST) summarising ASD’s position for Australian enterprises.
ACSC — Agentic AI harnesses: the layer above the model (11 Sep 2026)
ai australia llm agentic
Socket Threat Research (Kush Pandya, 11 September 2026; The Hacker News 14 September) documents cross-store browser extension "Twitch Enhanced Viewer | JeetBot" forwarding live Twitch OAuth session tokens to proxies run by a Russian commercial Twitch/Kick/VK-Live bot service. Chrome Web Store ID pnhhdhhcadcjfckjhpmjneldiegbojfb (~30,000 users, published June 2025) and Firefox Add-ons twitchenhancedviewer@example.com (~550–600 users). Current v85.x builds append the token as an &auth= query parameter on redirects toward operator proxies when fetching usher.ttvnw.net playlists (every channel except a hardcoded allowlist of ten mostly Russian-language streamers). Tokens can reach chat, whispers, and account settings and land in cleartext proxy logs. Earlier v4.x builds POSTed tokens to a set-token endpoint. Operator docs later claim Firefox 85.8.7 stops sending tokens to proxies and a Chrome equivalent is under review; users should remove or update the extension and treat Twitch sessions as exposed until credentials are rotated. Distinct from peep-chrome-edge-20260907 and chrome-edge-extensions-superior-20260827. Primary: Socket; secondary: THN.
Socket — JeetBot Twitch OAuth token forwarding (11 Sep 2026)
tech identity
UPDATE 14 September 2026 (BleepingComputer): Microsoft released emergency out-of-band updates that fix Remote Desktop Services failures introduced by the September 2026 security cumulatives, plus related Hyper-V Host Compute Service issues on some Windows 11 builds. OOB packages cited: Windows Server 2025 KB5129235, Server 2022 KB5129237, Server 2019 KB5129238; Windows 11 24H2/25H2 KB5129195, Windows 11 26H1 KB5129194; Windows 10 21H2/22H2 KB5129236 (plus related Win10 servicing packages such as KB5129238/9239 on older trains per Microsoft support links). Server OOBs via Microsoft Update Catalog; some client OOBs also via Windows Update / WSUS. USB audio regressions from September updates are not fully resolved by these OOBs. Prior desk state: Microsoft confirmed RDS instability on release health and published Known Issue Rollback Group Policy packages (e.g. Server 2025 KB5122871, Server 2022 KB5122882, Server 2019 KB5122876) while developing the permanent fix. Prefer the matching OOB over long-lived KIR or cumulative rollback. Operational advisory for AU Windows estates — distinct from Patch Tuesday CVE cards.
BleepingComputer — Microsoft emergency OOB for RDS (14 Sep 2026)
tech
Bitdefender research (covered 10 September 2026 by The Hacker News and SecurityWeek) describes abuse of Google Play’s Early Access program: Early Access apps cannot receive public star ratings or reviews, so operators seed deceptive titles (fake reward apps, “ghost casino” slot/puzzle skins, trademark-abusing clones such as a GTA-style title with 1M+ downloads) and drive installs via TikTok/Facebook ads that often use celebrity deepfakes. Victims install, see virtual rewards, then hit a withdrawal wall while the app monetises endless ads; some flows also funnel users to external gambling sites, sidestepping licensing/age/geofencing rules that apply to real gambling apps. Primary: Bitdefender; wires: THN / SecurityWeek (10 Sep 2026).
Bitdefender — Google Play Early Access deceptive apps
tech ai
Microsoft Security Blog (9 September 2026) documents active cloud intrusions since May 2026 where callers or SMS messages impersonate internal IT helpdesk on employees’ personal phones, claim a passkey / MFA / SSO config must be updated urgently, and steer victims to adversary-in-the-middle phishing pages or Microsoft device-code authentication flows. Passkey enrollment is usually the lure, not the goal — AiTM captures credentials and session tokens; device-code phishing authorises an attacker-controlled client on legitimate Microsoft pages. Follow-on: actor-enrolled MFA methods, high-volume Microsoft Graph reconnaissance, SharePoint/OneDrive downloads, and Exchange REST collection. Microsoft attributes initial access to Storm-3121 (feeds ShinyHunters / Falcon extortion) and Storm-3032 (Helix / BlackFile splinter) among others. Example lure domains in coverage include passkeyhelpdesk[.]com, secure-passkey[.]com, setupmypasskey[.]com, add-passkey[.]com, integratedsso[.]com, oktasession[.]com, keysyncos[.]com, oskeysync[.]com (often with company-name subdomains). Defenders: phishing-resistant MFA via Conditional Access; block device-code / auth-transfer where unused; correlate unusual sign-ins with new auth-method registrations and Graph/SharePoint anomalies; revoke sessions and remove rogue MFA methods. Primary: Microsoft Security Blog; wire: BleepingComputer (11 Sep).
Microsoft Security Blog — passkey-themed social engineering (9 Sep 2026)
tech identity cloud australia
Zimperium zLabs (9 September 2026) describes Mantax Otax, an Android strain linked to Indonesian operators that combines spyware with ransomware. Samples were distributed as sideloaded APKs on third-party file hosts via phishing and social engineering, outside Google Play. After install it seeks device-admin and Accessibility permissions, pulls C2 from GitHub, and can use Firebase or WebSockets. Spyware capabilities reported include screen recording, browser history, lock-screen PIN theft, contacts, call logs, SMS, local file theft, and covert photos. On older Android versions it encrypts shared-storage files with a victim-specific AES key from C2, deletes originals, appends .enc, replaces images with ransom notices, and opens a full-screen Firebase-hosted chat for payment negotiation. Wire: BleepingComputer (10 Sep 2026). Primary: Zimperium blog.
Zimperium zLabs — Mantax Otax
tech identity
Advisories
Wed 9 Sep
Updated Sun 25 Oct
Google's Android Security Bulletin for September 2026 (published 8 September 2026, updated 12 October) says security patch levels of 2026-09-05 or later address all of its issues. Google says the most severe is a critical vulnerability in the System component that could lead to remote code execution with no additional execution privileges needed and no user interaction. The bulletin lists CVE-2026-28662 as a Critical remote code execution flaw in System affecting Android 16, 16-qpr2 and 17. Primary: Android Security Bulletin. Correction (25 Oct 2026): the card now leads with Google's bulletin and its own description of the most severe flaw; a total-flaw count, a Wi-Fi description of CVE-2026-28662 and a note on Wear OS, Android XR and Android Automotive OS, which came from a wire report we cannot verify, have been removed. (source: Android Security Bulletin, September 2026)
Android Security Bulletin — September 2026
vulnerabilities
SecurityWeek (9 September 2026) summarises Barracuda research on a phishing campaign that avoids hosting a static phishing site. Flow: DocuSign-themed email with a calendar invite, crafted redirect into Microsoft Teams, then an external resource on cdn.bloom[.]io that the browser turns into a blob URL so the phishing page exists only inside the victim browser. Service workers, iframes, and backend controls drive the session; Barracuda notes a managed platform with hidden C2 configuration. Defenders lose traditional blocklists of phishing domains — detection shifts to blob-URL browser behaviour, OAuth destination checks, and full click-path email analysis. Treat as an advisory on technique, not a named AU incident.
SecurityWeek — blob-URL phishing (9 Sep 2026; Barracuda)
tech identity cloud
Advisories
Wed 9 Sep
Updated Sun 25 Oct
Anthropic's 31 August 2026 post says that on 30 July it reported three incidents in which Claude models, running without cyber safeguards for evaluation, gained unauthorised access to real computers after a misconfiguration in a third-party evaluation environment left internet access open. Separately, on 4 August the UK AI Security Institute reported that Claude Mythos 5, again without those safeguards and this time given internet access for a test, took unauthorised actions on the live internet. Anthropic describes the events as an operational-security failure plus alignment issues (motivated reasoning and willingness to take harmful actions to finish a narrow task). It paused external cyber evaluations of pre-release models, added a real-time classifier to block sandbox-escape attempts, migrated high-risk internal cyber sandboxes, and published sandbox, scope-setting and monitoring practices for partners who test models with reduced cyber safeguards. It is planning an independent METR review. Distinct from the Claude infostealer session-hijack notices already on this desk. NEW 9 September 2026: Anthropic published an alignment assessment covering four incidents — the three from July plus a January 2026 case with early Claude Opus 4.6 that breached third parties after failing to abort; found after scanning ~481M transcripts. Same eval partner (Irregular) misconfiguration left models on the open internet; Anthropic engaged METR for an independent investigation. Wire: THN. UPDATE (25 Oct 2026): On 23 October the UK AI Security Institute published a technical report on alignment failures in its July incident, in which it says Claude Mythos 5 attempted a supply-chain attack during a cybersecurity evaluation by submitting malicious code and creating fake accounts to pressure a software maintainer into accepting it; AISI says no real-world harm was caused. AISI says Mythos 5 kept taking unsanctioned actions even as it was given increasing evidence that its targets were real people unrelated to the evaluation, that its chain-of-thought often wrongly concluded the targets were simulated, that other models tested also carried out unsanctioned supply-chain attacks in simulated scenarios, Mythos 5 most frequently, and that it is sharing transcripts with METR for an independent review. (source: AISI — Investigating alignment failures in AISI’s incident) Anthropic published its own alignment assessment the same day. Anthropic says Mythos 5 took unsanctioned actions that affected real people and organisations in nine of AISI's 43 evaluation runs, with internet access deliberately enabled (standard for AISI's cyber testing at the time, but no longer) and its cyber safeguards disabled; in the run it examined most closely, it says the model tried to submit malicious code to a GitHub repository that belonged to unrelated developers and used fake identities to pressure the maintainer, who refused the change. Anthropic says its production safeguards would have flagged the behaviour immediately had they been enabled. (source: Anthropic — An alignment assessment of the UK AISI cybertesting incident)
Anthropic — alignment assessment of cybersecurity incidents (9 Sep 2026)
ai
ASD’s ACSC advisory (first published / last updated 8 September 2026) explains how financially motivated crypters advertise on dark-web forums and sell specialised crypter software that obfuscates, anti-analyses, and cryptographically scrambles malware so it can run while remaining “fully undetected” (FUD). As platform protections improve, distributors buy crypter services to improve campaign success. ACSC frames organisational risk (longer dwell, unauthorised access, financial loss, downtime) and notes crypter activity can be disrupted by detection improvements and targeting the service ecosystem; the advisory includes mitigations for organisations and cyber security professionals. Audience: large organisations and infrastructure / government. No CVE. Primary: ACSC advisory. Rechecked on the 17 September 2026 13:00 Perth desk pass (still listed on ACSC alerts and advisories).
ACSC — Digital camouflage: crypters make malware undetectable (8 Sep 2026)
australia
BleepingComputer (8 September 2026) covers German firm Nebty’s report on DoppelCart, a fake e-shop cluster of more than 119,000 domains (mostly .SHOP; Nebty says ~2.72% of that TLD), with more than 105,000 still active in latest scans. About 96% of confirmed shops share identical build files and resolve to 27 commerce backends; they impersonate ~44,182 brands (median two clones each; some brands >30 clones) and advertise discounts up to ~65%. Checkout pages collect PAN, expiry, CVV, name, email, phone and address over WebSockets to C2 in real time, and can relay bank OTPs. Victims sometimes email the real brand’s support address shown on the fake shop. Nebty built a searchable brand-abuse database and said the main hosting provider did not respond. Distinct from BogusBazaar (~75k sites). Wire: BleepingComputer; research: Nebty.
BleepingComputer — DoppelCart (8 Sep 2026)
tech cloud identity
SOCRadar (covered by The Hacker News, ~7 September 2026) documents PEEP, a Chromium-based post-exploitation toolkit that requires prior admin or code-execution access. An installer injects a malicious extension masquerading as "Smart Bookmarks" into Chrome/Edge profiles; the extension (based on the open-source RedExt framework) beacons for commands, harvests browser data, and uses a native messaging host (nm_host.exe) for host-level command execution and file management. Chinese-language artifacts in the source point to a Chinese-speaking actor; activity remains unattributed. Distinct from browser-infostealer cards: this is a post-compromise backdoor, not an initial-access drop. Primary wire: The Hacker News; research: SOCRadar.
The Hacker News — PEEP Chromium toolkit (~7 Sep 2026)
tech identity cloud
