Weekly recap · Security Now #1094
31 August – 6 September 2026
SonicWall SMA 1000 zero-days, ShinyHunters claims 284M McKesson patient records, infostealers hijack Claude sessions, and CISA drops six free assessment programs.
Top stories
- SonicWall SMA 1000 zero-days exploited
#vulnCRITICALZERO-DAYKEV— Attackers chained CVE-2026-83548 (SSRF) and CVE-2026-83549 (OS command injection) on SMA 1000 appliances before patches existed. Why it matters: yet another remote-access appliance used as the front door. Help Net Security - ShinyHunters claims 284 million patient records from McKesson
#breach— McKesson disclosed that attackers got into third-party applications and stole data. Why it matters: potentially one of the largest healthcare data thefts ever, via SaaS rather than core systems. Help Net Security - Anthropic locks out Claude users after infostealers hijack sessions
#threat#ai— Stolen session cookies from infostealer infections were used to take over Claude accounts. Why it matters: AI assistant accounts now hold code, documents and credentials worth stealing. Help Net Security - ~22,000 Exchange servers still exposed to CVE-2026-62911
#vulnCRITICAL— A critical authentication bypass remains unpatched on tens of thousands of internet-facing servers. Help Net Security
Breaches & incidents
- Thomson Reuters C-Track breach exposes US and Canadian court records
#breach— Help Net Security
Vulnerabilities & patches
- AI-stack KEV additions: LiteLLM, Starlette, Kestra, JFrog Artifactory
HIGHEXPLOITEDKEV— CISA added CVE-2026-59822 (LiteLLM auth), CVE-2026-48710 (Starlette request smuggling), CVE-2026-49869 (Kestra command injection) and CVE-2026-82329 (Artifactory auth) on 2 Sep. AI gateways and MLOps tooling are now routine targets. CISA KEV - Sangoma Switchvox SQL injection exploited (CVE-2026-9586)
HIGHEXPLOITEDKEV— Help Net Security - Chrome V8 type confusion CVE-2026-85046
HIGHZERO-DAYKEV— Added to KEV on 4 Sep; later shown to be part of a Chrome→Windows chain used by Chinese actors (see 21–30 Sep). CISA KEV - PaperCut servers backdoored with legit remote-access tools
EXPLOITED— Help Net Security - Patch Tuesday forecast: August was the second-biggest ever (398 CVEs)
PATCH— Help Net Security
Threat intel & red team
- Russian hackers plant “nuclear weapon” prompts in malware to trip AI analysis guardrails
#threat#ai— An anti-analysis trick aimed at defenders who use LLMs to triage samples. Help Net Security - Fake “Claude Opus 5” GitHub repo delivers RevStealer
#threat— Help Net Security - “Spring Ring” vishing via Microsoft Teams
#threat— Fake IT-support accounts call employees on Teams to gain a foothold. Help Net Security · Microsoft Security - FBI warns of OAuth consent phishing against prominent people and their families
#threat— Help Net Security - Inside an AI-assisted intrusion
#threat#ai— Unit 42’s investigation of an attack driven by autonomous AI tooling. Unit 42 - Attackers pose as AI crawlers to find exposed credentials
#threat— Help Net Security
Blue team & defense
- ASCII smuggling moves from prompt injection to phishing evasion
#defense— Invisible Unicode tag characters hide content from filters. Microsoft Security - Counterfeit installers → full compromise
#defense— Hunting guidance for a fake-download campaign. Microsoft Security - Sift: open-source secrets scanner for M365, Slack and Jira
#defense— Help Net Security
Industry, policy & developments
- CISA ends six free assessment programs for critical infrastructure
#industry— Including Cyber Resilience Reviews and Ransomware Readiness assessments; operators are pointed to the Cross-Sector Performance Goals instead. Tech Times - CISA makes the case for eliminating whole vulnerability classes
#industry— Help Net Security - NIS2: fix IAM and access control before your audit
#industry#eu— Help Net Security - Debian rejects an LLM ban; AI disclosure stays voluntary
#industry— Help Net Security
Worth reading
- SpecterOps Skills: demystifying agent tradecraft
#learning— SpecterOps - What ingesting malware intel at scale really takes
#learning— Help Net Security
Security Now cross-check
SN #1094 — “AI Patching Shortcomings” (1 Sep)
- More implants found in Chinese-made ZBT routers
#threat— VulnCheck rooted a 2019-era ZBT router and found two older implants beyond the ENDLESSDOORS backdoor disclosed in August (CVE-2026-66747, 100K+ units). Steve’s advice: stick with name brands. The Hacker News · CSA research note · Transcript - Frontier models are poor at fixing the bugs they find
#learning#ai— Research exhaustively testing AI vulnerability remediation found it lags far behind discovery and exploitation. Transcript