Two themes emerge from today's briefing. The first is the confirmation that North Korea was behind the September 2025 debug and chalk npm hijack — packages with more than two billion combined weekly downloads — alongside Google Threat Intelligence Group's finding that malicious open source packages increased 1,444% between 2024 and 2025. The broader trend is the increasing scalability of software supply-chain compromise. Dependency relationships allow a single compromised component to propagate rapidly across large numbers of downstream applications, as demonstrated by the axios compromise in March 2026, which spread to tens of thousands of dependent packages within hours of publication.

The second theme is the cluster of Microsoft platform disclosures in a single day. A Copilot for Word prompt-injection worm, Laundry Bear's exploitation of CVE-2026-42897 in Outlook Web Access, and a patched Azure Cosmos DB sandbox escape affect different layers of the Microsoft ecosystem — productivity tooling, on-premises mail infrastructure, and cloud database services. Considered individually, each is a contained disclosure with its own remediation guidance. Together, they show how broad platform ecosystems create interconnected exposure across productivity tools, communication systems, and cloud services. The Copilot worm is particularly notable because it propagates through behaviour resembling normal user activity, leaving no conventional malware artefact for endpoint tools to identify.

The Toy Ghouls group deployed a custom ransomware called GenieLocker against Russian manufacturing organisations starting in March 2026. Attackers gained initial access via a compromised OpenVPN partner connection, then used Mimikatz, PsExec and RDP/SSH for lateral movement before encrypting files with XChaCha20-Poly1305 on Windows, Linux and ESXi systems. The Windows variant includes anti-debugging checks and a hardcoded "secret argument" required to execute, while no data exfiltration was observed.

Semiconductor manufacturer Analog Devices told the SEC it detected unauthorised access to its systems on 23 June, followed by data exfiltration of an unspecified scope. The company reported a separate, unconnected cybersecurity matter on 26 July, which it is still assessing. Ransomware group ExfilSquad has separately claimed to have stolen over 570,000 customer records, though Analog Devices has not confirmed a link.

CISA added CVE-2026-20316 to its Known Exploited Vulnerabilities catalog, citing evidence of active exploitation. The flaw is a hard-coded password vulnerability in Cisco Secure Firewall Management Center software. Cisco has released hot fixes for affected releases spanning versions 7.0 through 10.0.

A phishing campaign is impersonating India's Income Tax Department, sending fake penalty notices over WhatsApp that link to a malicious ITD.zip archive. Android victims receive spyware capable of intercepting OTPs and banking credentials, while Windows victims receive an executable disguised as svchost.exe that fetches a second-stage payload from an Alibaba Cloud bucket. The operation also includes lookalike e-Filing portals and fraudulent refund messages hosted on disposable, low-trust domains.

Kaspersky identified two previously unknown backdoors, OctLurk and SilkLurk, targeting government organisations across Central Asia since January 2025, likely operated by a Chinese-speaking actor. Both use DLL side-loading and victim-specific encryption keys derived from machine data to evade detection and load additional plugins for keylogging, credential theft and remote access. Attackers also deployed the PlugX remote access trojan and a proxy tool named LurkProxy, with infrastructure overlapping a separate Linux-focused malware campaign in Kazakhstan.

A researcher disclosed a cross-domain prompt injection vulnerability in Microsoft Copilot for Word that allows malicious instructions hidden in a source document to silently alter AI-generated output and copy themselves into downstream documents. Each affected document becomes a new carrier, enabling the attack to spread through ordinary editing workflows without any further involvement from the original malicious file. Microsoft deployed two mitigations over a 144-day disclosure period, but the vulnerability class remained exploitable at publication using modified payloads.

Wiz Research discovered a sandbox escape in Azure Cosmos DB's Gremlin API that allowed arbitrary code execution on the multi-tenant gateway service handling customer queries. From there, researchers accessed a platform-wide signing key, dubbed the Cosmos Master Key, which could retrieve the primary key of any Cosmos DB account across all tenants, regions, and API types. Microsoft deployed an initial hotfix within 48 hours of the November 2025 disclosure and completed a full architectural remediation in July 2026, with no evidence of exploitation found.

Google Threat Intelligence Group reports a 1,444% increase in malicious open source packages from 2024 to 2025, driven by large-scale campaigns targeting PyPI, npm, and Docker Hub. North Korean actor MIDNIGHT NEPTUNE compromised the widely used axios npm package in March 2026 via a maintainer account takeover, deploying the WAVESHAPER.V2 backdoor to victims across 15 industry verticals and 13 countries. A separate group, UNC6780, abused GitHub Actions pull request triggers to steal credentials and pivot into broader network environments, monetising access through ransomware partnerships.

Security researchers at Bitsight discovered that H96 Android TV streaming sticks contain pre-installed apps linked to Chinese firm Zhejiang Fengwo IoT Technology, which uses the devices to conduct ad fraud against AI-generated websites whilst simultaneously operating as residential proxies. The devices spoof their identity as mobile phones to click ads on sham websites that only display content to traffic matching that spoofed profile, with Bitsight estimating the operation generates close to $50,000 per day from roughly 38,000 tracked devices alone. The devices alternate between proxy and ad fraud activity based on whether a television is connected and active via HDMI.