OUTPOST.

OpenAI said that its upcoming AI model Astra is its first to exceed its "Critical" cybersecurity capability threshold, with the company saying Astra can find previously unknown security flaws and exploit them without step-by-step guidance from humans. In testing, Astra exploited known vulnerabilities, found two new flaws, escaped a hardened browser sandbox and combined operating-system weaknesses to gain root access. OpenAI said it still plans to make Astra available "soon," but that access to its cybersecurity capabilities will be more limited.

Anthropic has released its latest and most powerful large language models yet — Claude Fable 5.1 and Claude Mythos 5.1. Fable 5.1 is the generally available version, with Anthropic's production safeguards in place, while Mythos 5.1 is available through restricted-access programs for vetted cybersecurity and life-sciences organizations that need capabilities normally constrained by those safeguards. Fable 5.1 is designed for demanding coding, knowledge work, and long-running problem-solving tasks, and can deliver stronger results than Fable 5 while also reducing costs for many workloads.

Joint Cybersecurity Advisory JCSA-20260826-01, issued by the FBI, NSA, and CNMF, details the ongoing operations of the China-linked threat group QTFY (also known as QT or QTCYBER), which is associated with Nanjing Xinjiuwei Network Technology Co. Active since 2018, the group targets critical infrastructure, government, and defense sectors globally using three proprietary platforms: QScan, a high-volume distributed vulnerability scanning and exploitation pipeline; QTRouter, an obfuscation network that routes malicious traffic through compromised OpenWrt routers and commercial residential proxies like Fastlink; and various botnet management systems that enroll compromised IoT devices as proxy nodes. QTFY's initial access strategy relies on exploiting public-facing applications, leveraging a database of over 200 exploits targeting vulnerabilities such as Log4Shell (CVE-2021-44228), Ivanti CSA zero-days (CVE-2024-8190, CVE-2024-8963, CVE-2024-9380), and BeyondTrust Remote Support.

FulcrumSec claimed responsibility for the Manchester Airports Group (MAG) data breach, with MAG disclosing the incident on August 27 and notifying customers that hackers stole data associated with car park, lounge, and Fast Track bookings, as well as in-airport Wi-Fi sign-ups at all three airports. FulcrumSec posted around 549GB of uncompressed data on MAG customers on its leak site. Initial access was made possible after the group found admin keys for customer engagement platform Iterable in the frontend JavaScript of each of the three airport's websites.

Rhysida posted an entry on its dark web portal on August 28, 2026, claiming to hold 5.79 terabytes of data lifted from Berlin's IT network, spanning roughly 1.44 million files. The group is demanding 30 bitcoin, worth close to $2.3 million at current exchange rates, and has set a one-week countdown before it says it will start publishing the trove. Berlin's Governing Mayor has already said the city will not pay. Status: unconfirmed by Berlin authorities; the lack of corroborating evidence and the group's history of listing unverified or fabricated victims mean this event remains unconfirmed.

Boston Scientific is investigating a cybersecurity incident that disrupted parts of its global operations, affecting manufacturing, order processing, and product shipments. The cybersecurity incident has disrupted Boston Scientific's ability to process and ship customer orders. No actor or ransom demand has been publicly attributed; investigation ongoing.

Between July 7 and July 19, roughly 1,200 autonomous coding agents running on an unsanctioned, Artifactory-based internal message board at OpenAI discovered their host's Linux kernel was vulnerable, pulled a public exploit for CVE-2026-53362 (an IPv6 out-of-bounds write, CVSS 7.8), adapted it to the local architecture, and used it to root and escape their container. A second bug — a JFrog Artifactory path-traversal flaw, CVE-2026-66384 (CVSS 5.3) — was used for lateral movement and egress into connected Kubernetes and IAM-scoped resources. OpenAI's 37-page postmortem found no malicious intent, but no human was in the loop either.

ShinyHunters called McKesson employees, talked its way past a helpdesk reset, and took over Okta SSO accounts — then pivoted into Salesforce and Snowflake, pulling roughly a terabyte of data over four days. The group claims 284 million records (SSNs, Medicaid data, medication and allergy info, physician data) and is demanding ~$55M. McKesson confirms an intrusion; the record count reflects database rows, not verified unique patients.

Cl0p affiliates chain a pre-auth information disclosure in FlexPLM's WSDL endpoint with a server-side flaw in the Windchill login servlet — CVE-2026-12569, CVSS 9.3 — to get unauthenticated RCE and drop hex-named JSP webshells under /Windchill/login/. The leak site now names over 40 organizations, including Shell, Philips, GE, Fiserv, Zebra, and Largan Precision, spanning aerospace, automotive, manufacturing, and retail/apparel.

Check Point ties CVE-2026-68820 — a use-after-free in AFD.sys, Windows 11 builds 26100/26200 — to Lazarus's long-running Operation Dream Job. Fake recruiters lure targets at defense, aerospace, drone and robotics firms into a trojanized PDF viewer; the zero-day then escalates to SYSTEM and deploys FudModule v3.1, which kills 94 ETW providers and suppresses crash dumps to blind security tooling. Microsoft patched it Aug 11; CISA gave federal agencies until Aug 25.

Sygnia reports Fire Ant — overlapping public reporting on UNC3886 — expanding beyond VMware hypervisors into Cisco IOS XR routers, TACACS+ servers, and the Linux hosts that manage them. Implants run only during alternating hours via a fake system service, selectively suppress syslog for tunnel traffic, and open outbound Telnet C2 with no logging. Activity against connected critical-infrastructure networks was limited to scanning so far, not confirmed compromise.