Governed Hack-Back, Cloud AI Updates, and Evolving Threat Activity

Policy, cryptography, and platform updates intersected with active threat activity and patch guidance. A new federal framework outlines conditions for private-sector participation in limited offensive cyber operations, while hyperscalers advanced AI availability and cryptographic roadmaps. Concurrently, researchers detailed multifaceted espionage and fraud campaigns, ransomware tradecraft, and Windows/SCCM hardening steps.

Federal Oversight for Limited Hack-Back Operations

The White House published a National Security Presidential Memorandum establishing a program to authorize vetted private firms to conduct limited offensive operations against foreign criminal groups under federal supervision. The memorandum directs the National Coordination Center to manage participation, with oversight by executive directors from the Departments of Justice and Homeland Security. Participating companies must be vetted, contract with DOJ or DHS, maintain a $1 million bond or escrow, and immediately halt activity if it strays beyond approved parameters or risks affecting U.S. persons or systems. The framework emphasizes constitutional, legal, and international compliance and introduces rigorous review and oversight procedures. Coverage in BleepingComputer notes mixed industry reactions, with supporters citing mounting consumer losses and critics warning of attribution errors and operational risks.

Cloud AI and Productivity: New Models and Integrations

AWS Bedrock now offers governed access to OpenAI’s Daybreak Blue (GPT-5.6 Sol) and Daybreak Red (GPT-5.6 Cyber) for eligible customers in US East (N. Virginia). Blue is positioned for vulnerability discovery, detection engineering, and incident response, while Red supports advanced, authorized research such as exploit reproduction and mitigation development under stricter identity verification, monitoring, and access controls. Both run on Bedrock’s next-generation inference engine with zero-operator access enforced at the chip level, and customer inference data is not used for training. Access requires enrollment through OpenAI and AWS approval.

AWS GovCloud added Claude Opus 5 via Amazon Bedrock, with zero data retention enabled by default. The model targets long-running agentic workflows, improved navigation of large codebases, production-quality code generation, and enhanced reasoning over lengthy, document-heavy inputs. Availability spans both GovCloud regions through bedrock-runtime and GovCloud (US‑West) via bedrock-mantle, enabling regulated organizations to maintain data residency and governance.

Amazon Quick announced general availability of Microsoft 365 extensions that embed its AI capabilities into Excel, PowerPoint, Word, and Outlook. The Excel extension focuses on analysis, pivots, charting, and data cleanup; PowerPoint automates slide creation and refinement using organizational templates; Word supports formatted document generation and large edits with track changes; Outlook assists with prioritization, inbox organization, scheduling, and drafting replies using Quick data and inbox context. The rollout targets operational gains across finance, sales, marketing, legal, operations, and IT.

Cryptography Shifts: Post-Quantum Plans and Certificate Changes

Google Cloud published a phased roadmap to migrate to post-quantum cryptography across three risk domains. The plan aims to mitigate store-now-decrypt-later risk by end of 2027, harden digital signatures, and rebuild key management for cryptographic agility through 2028, with some hardware-dependent elements potentially extending past 2029. Shipped capabilities include hybrid ML-KEM for major API endpoints, opt-in hybrid TLS 1.3 on load balancers, and Cloud KMS GA for ML-KEM, ML-DSA, and SLH-DSA. Google is pursuing Merkle Tree Certificates to address large signature performance, and expects customer-side updates for negotiating quantum-safe handshakes and key lifecycle management.

AWS blog announced that AWS Certificate Manager (ACM) will discontinue email-validated public certificates by September 30, 2027, aligning with the CA/B Forum’s deprecation effective March 15, 2028. Customers are advised to migrate to DNS validation; ACM’s UpdateCertificateOptions API supports in-place switching so certificate ARNs remain unchanged. During a 72-hour DNS record window the certificate continues to function on email validation, and once DNS validation completes, ACM is designed to auto-renew without further manual steps.

Complementing that policy shift, AWS What’s New details new ACM support for changing domain validation from email to DNS without reissuing certificates or changing ARNs. ACM will stop issuing email-validated certificates on March 31, 2027, and cease renewing them on September 30, 2027. Customers receive per-domain CNAMEs, can monitor validation status via console or API, and are advised to prefer DNS validation (or HTTP validation for CloudFront distributions) for automated renewals and continuity across pipelines and load balancers.

Threat Activity, Breaches, and Patch Watch

Symantec research covered by BleepingComputer documents Jewelbug (also tracked as Earth Alux / REF7707) combining espionage against government and military targets with large-scale cryptocurrency fraud. The group injected a webmail template that exfiltrated cookies and addresses via WebSocket, used a fake Flash update to deploy the Antino backdoor, and leveraged tooling including XG‑Web and a malicious PDF Viewer extension. Logs indicated more than one million implant check-ins, over 580,000 cookies stolen, and thousands of credentials and email bodies exfiltrated, with focus across the Middle East, Southeast Asia, and South Asia. In parallel, the group operated a CMS fleet and lookalike domains driving traffic to fake exchanges and other lures.

Researchers chronicled an AI agent–assisted, near-autonomous campaign against Asian government infrastructure that used open-source agent frameworks across 12 waves to automate reconnaissance, credential attacks, and lateral movement. The operation generated thousands of exfiltrated personnel records and dozens of cracked credentials and expanded to supply chain partners and critical sectors. The report highlights use of Hermes and OpenClaw agents, evidence of a DeepSeek‑V4‑Flash model, and exploitation of unauthenticated APIs and token validation flaws, while avoiding specific actor attribution. Details appear in CSOonline.

An Akira ransomware affiliate used valid VPN credentials on a SonicWall device lacking MFA to move from initial access to exfiltration within about five hours. The intruder forced Safe Mode with Networking to disable most security agents temporarily, installed AnyDesk for persistence, and attempted encryption that failed due to low virtual memory and PowerShell errors. Despite failed encryption, data theft and credential exfiltration enabled extortion. Huntress-reported findings in BleepingComputer recommend enforcing MFA on VPNs and monitoring for Safe Mode boot configuration and Safe Mode registry persistence.

CRM provider Beacon attributed a breach affecting roughly 1,500 UK charities to a compromised AWS access key likely exposed in public JavaScript build artifacts. Valid credentials enabled downloads of the platform’s dataset during a 1.5‑hour window, with AWS reports showing correlated activity. Exposed data likely includes supporters’ contact and donation details but not sensitive patient records or payment card/bank information. Impacted organizations were urged to report to the UK ICO and remain alert for social engineering. Coverage by Infosecurity notes reset credentials and no signs of persistence.

The UK ICO reprimanded the Criminal Records Office (ACRO) over a 2023 incident exposing 10,920 individuals’ highly sensitive data via unauthorized access to its website and Kentico CMS. Failures included ineffective patch management and unreviewed malware alerts from a deployed security solution, while segmentation limited scope and remediation followed. The reprimand underscores governance and oversight alongside technical controls. Full details appear in Infosecurity.

On enterprise tooling, CSOonline reports an XM Cyber chain against Microsoft System Center Configuration Manager that can escalate a non‑privileged domain user to SYSTEM on the primary site server using multiple weaknesses: an AdminService authorization bypass (patched in July as CVE-2026-47301), a CabSlip path traversal, lax signature validation accepting inexpensive or leaked certificates, and an unsigned DLL load path in SMS Executive. Microsoft’s patch closes the initial bypass for standard users; exposure remains for certain privileged roles. Recommended defenses include restricting AdminService access, auditing RBAC, and monitoring for adsource.dll changes.

Microsoft addressed a local privilege escalation zero‑day dubbed LegacyHive (CVE-2026-62832) in August Patch Tuesday. The flaw in the Windows User Profile Service’s link following enables loading and modifying another user’s registry hive to gain administrator privileges when additional credentials are available. Researchers validated the public exploit and released hunting queries, and ACROS previously issued temporary patches. See BleepingComputer for remediation context.

Cisco Talos detailed JWR, a live-operator phishing framework that uses WebSockets to observe keystrokes in real time and tailor fake checkout and login flows, capturing payment data, 2FA codes, identity documents, and device fingerprints. SMS lures impersonate regional toll and postal authorities, and real-time interaction can defeat conditional access controls by harvesting tokens when needed. Talos recommends user education on smishing and phishing‑resistant MFA. Indicators and coverage are provided in Talos.

Check Point’s Q2 2026 ransomware analysis found 2,139 leak-site victims, flat quarter‑over‑quarter but up 33% year‑over‑year, alongside broader mid‑tier competition with 93 active groups. Leaked logs from The Gentlemen illustrated AI‑assisted rapid tooling development and a 90/10 affiliate split. With payment rates around 23% yet substantial 2025 on‑chain totals, groups are leaning into data theft‑first extortion. The report emphasizes equal priority on initial access, exfiltration detection, and exposure reduction. Mitigation guidance appears in Check Point.

Separately, hardware wallet maker Trezor disclosed a breach at logistics provider ShipMonk affecting nearly 14,000 customers who received shipments between May 10 and August 8, 2026, exposing order-related contact details. ShipMonk attributed access to a zero‑day exploited in the Metabase analytics platform and reported a patch and session invalidation. Trezor warned of increased phishing and impersonation risks. Details are in BleepingComputer.