Cyber Digest

Executive Summary

A quiet Monday opens the new week — most outlets did not publish over the weekend window beyond the coverage already captured in Saturday's and Sunday's digests, and no major new zero-days, endpoint breaches, ransomware events, or KEV additions emerged this morning. First, the freshest first-party publication in the window is ACSC's guidance "When AI agents take unexpected actions", published 14 August, walking Australian organisations through how AI agents can take unanticipated actions, the risk of goal misalignment, and practical safeguards for safer adoption. It is the latest step in a sustained ASD/ACSC push on AI-agent security this month, and it anchors the only genuinely new first-party output of the cycle — no new ACSC alert or advisory has been issued since the CMS and Zimbra advisories of July. Second, in a substantive consumer-data and adtech piece (14 August), Brian Krebs profiles a new free service called DecryptAds that scrapes and cross-correlates the `ads.txt` / `app-ads.txt` / `buyers.json` / `sellers.json` files websites and apps publish, making it easy to see exactly which adtech firms and data brokers are tracking users — and surfacing that major outlets (including top US military-news sites and ESPN) work with ad-network entities flagged as based in China, Russia, the UAE, Cyprus or Panama, some processing payments through Russian banks placed under Western sanctions. Third, the only genuinely fresh weekend item with a sector angle is an FBI request for truck drivers nationwide to come forward as potential victims in a Georgia tax-preparer fraud case: prosecutors allege a Walton County bookkeeping firm collected payroll and estimated tax money from clients — most of them self-employed truckers — without remitting it to the IRS, in a 54-count indictment spanning 32 bank-fraud and 14 payroll-tax charges. Everything else in the fixed rotation — the SAP Commerce Cloud CVE-10.0 exploitation attempts, the Metabase KEV add, the macOS screen-sharing miner campaign, the Mustang Panda rootkit, JewelBug's dual-track operations, the Infoblox expired-domain campaign, the Scottish prosecutor's-office breach, and the US and German legal developments — was already covered across the 14–16 August digests and is treated here as context rather than padded re-posts.

Two threads from the past seven days frame this quiet Monday, grounded in the digests of the past week. The patch-to-exploitation conveyor continues to define operator risk. The week's slate — the SAP Commerce Cloud CVE-2026-58231 (CVSS 10.0) moving to attempted exploitation within days of the fix, the Metabase SQL-injection flaw added to CISA's Known Exploited Vulnerabilities catalogue, the macOS Screen Sharing flaw exploited on internet-exposed Macs to plant a Monero miner, and the SharePoint authentication bypass — kept CISA's KEV and the two-week BOD 26-04 federal patch directive central to all Five Eyes defender guidance. AI and AI-agent security is now the most consistent shared thread across allied agencies. This week saw ASD/ACSC's AI-agent guidance, NCSC-UK's 4 August CTO statement on frontier-AI evaluation incidents, widespread reporting on near-autonomous AI attacks (including the reported first near-autonomous AI attack on a Taiwanese government target), Kaspersky and Symantec research on AI-driven tooling, and the AI supply-chain incident arc (ChatGPT/OpenAI/Anthropic breaches of third-party AI platforms) all reinforcing the same message: adversarial capability is being compressed, and agentic systems are becoming both a target and a lever. The Krebs/DecryptAds investigation adds a supply-chain-vigilance angle that echoes the week's infrastructure-abuse theme — pointing to ad networks processing payments through sanctioned Russian banks operating on major western media properties — and connects to prior coverage of AI-generated "slop" websites and the H96 streaming-stick/Fengwo Group proxy farm. China-linked espionage continued its momentum across the week (Mustang Panda's signed kernel rootkit, JewelBug's dual state-espionage-and-crypto-track operations, PATCHCORD's Afghan telecom and Indian critical-infrastructure targeting), extending an Indo-Pacific-focused collection push that allied states are answering with their own offensive and commercial-policy moves (the US private-sector offensive-operations memo, Germany's spy-law overhaul, KOSA's advancement). Looking ahead, expect the AI-agent governance wave to produce more allied guidance in the coming fortnight, continued urgency on internet-facing commerce and remote-access patching, sustained China-linked APT activity across South and Southeast Asia, and a genuinely busier news cycle once US and European outlets resume their Monday publication runs.