AWS Tightens Controls; CISA Flags Windows IKE Exploits, PLC Threats
Coverage: 19 Aug 2026 (UTC)
< view all daily briefs >AWS delivered a slate of security and governance updates spanning identity enforcement, policy controls, private connectivity, and web-grounded AI features. The cloud provider also expanded isolation-focused compute and regional coverage for managed ingestion. Meanwhile, U.S. agencies warned of active exploitation targeting Windows IKE services and AI-assisted probing of Siemens PLCs, as ransomware activity and large breaches continued to pressure critical sectors. Open-source stakeholders outlined plans to coordinate responses to AI-enabled vulnerability reports at scale.
Identity, Governance, and Least-Privilege Controls
Amazon SageMaker notebooks now support Trusted Identity Propagation (TIP) with Amazon Athena, Amazon Redshift, and Amazon EMR Serverless, forwarding each user’s IAM Identity Center identity to AWS Lake Formation so queries execute under that individual’s permissions. This enforces table-, column-, and row-level boundaries and improves audit attribution by recording the specific user in CloudTrail. TIP reduces administrative overhead by eliminating extra logins or token management and is available wherever SageMaker Unified Studio is offered.
AWS also raised the default limit of managed policies attached to an IAM role from 10 to 20 via AWS IAM, reducing friction for least-privilege designs that rely on smaller, purpose-scoped policies; customers can request increases up to 25 through Service Quotas. Complementing that, Amazon Quick added a deny-by-default setting for custom permissions so new AI capabilities remain restricted until explicitly enabled, allowing administrators to gate adoption by category and maintain consistent control at account level.
Private Connectivity, Isolation, and Regulated Workloads
Storage Gateway now supports FIPS 140-3 validated endpoints over AWS PrivateLink for Tape and Volume Gateways, keeping traffic on the AWS private network and aiding compliance for regulated workloads. Operators create a FIPS interface VPC endpoint and enable the FIPS option during activation on gateways running version 3.2.7 or later. For public sector and regulated users, OpenSearch Ingestion is now available in AWS GovCloud (US-East and US-West), providing a fully managed, autoscaling ingestion tier with filtering, redaction, transformation, and routing for Amazon OpenSearch Service clusters and serverless collections.
Lambda MicroVMs expanded to five additional Regions—Mumbai, Singapore, Sydney, Frankfurt, and Stockholm—bringing VM-level isolation, near-instant launch/resume, and state preservation to more geographies for latency and residency needs. In Europe, AWS added a fourth Availability Zone in London, enhancing fault isolation and capacity for general compute and accelerated AI/ML instances such as Trn3 and P6 within AWS London AZ.
Web-Grounded AI With Tunable Source and Data Egress Controls
AWS introduced an external_web_access parameter in Bedrock’s Web Search, allowing administrators to control whether the service fetches content directly from the public web or confines retrieval to Amazon’s in-AWS web index and knowledge graph. Enabling external fetches requires the bedrock-websearch:ExternalWebAccess permission; the feature debuts in three U.S. Regions. The change, announced via Bedrock Web Search, lets teams trade off freshness for stricter data residency when needed.
For finer-grained sourcing, AWS added per-request domain include/exclude lists and published-date range filters to AgentCore’s Web Search tool, and expanded availability to Europe (Ireland) and Asia Pacific (Tokyo). These controls increase developer flexibility while keeping gateway-level allowlists and expanded domain caps for governance, as outlined in the AWS Bedrock update.
Separately, Amazon Bedrock now offers SpaceXAI’s Grok 4.6, a large language model with a 500K token context window and selectable reasoning effort levels for balancing computation and depth. The integration, detailed in Grok 4.6, provides enterprises with AWS-grade security, monitoring, and regional scale-out for complex, multi-step, and visual workloads.
Active Threats, Exploitation, and Enforcement Actions
CISA confirmed in-the-wild exploitation of a critical Windows IKE Service Extensions vulnerability (CVE-2026-33824) that enables unauthenticated remote code execution via specially crafted packets to UDP ports 500/4500. Microsoft patched the issue in April 2026. CISA advised prioritizing updates or, if patching is not immediately possible, filtering inbound traffic as interim mitigation. Details are summarized in Windows IKE RCE.
U.S. agencies (NSA, CISA, FBI, DOE, EPA) warned of AI-assisted attacks targeting Siemens S7 PLCs across critical infrastructure sectors. Adversaries use scanning services to find exposed devices and custom Python tools leveraging snap7 libraries to read/write memory and logic over S7comm, often disguised as legitimate OT tools. Mitigations include removing internet exposure, installing vendor updates, tightening access controls, and monitoring for anomalous PLC activity. See the joint alert coverage: Siemens PLC Alert.
In ransomware activity, the FBI, CISA, and HHS reported that Medusa ransomware has impacted more than 500 critical infrastructure organizations, with rapid weaponization of disclosed vulnerabilities and expanded tradecraft for initial access and exfiltration. Agencies urge rapid incident scoping, removal of C2 and RMM tools, credential rotation, patching the initial CVE, and use of CISA’s eviction guidance. Summary: Medusa Ransomware.
The U.S. Department of Justice unsealed charges against eight additional alleged members of Iran-linked Mabna Institute for a long-running campaign stealing academic and corporate data, bringing total indictments to 17. The operation targeted thousands of professor accounts and exfiltrated tens of terabytes of content valued at billions of dollars. Rewards for information are being offered. Coverage: DoJ Charges.
Japan’s Sakura Internet reported unauthorized access to its sales management system following an earlier breach of its rental server service. Up to 1,360,563 member accounts may be affected; passwords are hashed and no credit card data is stored in the impacted system. The company invalidated abused credentials, removed malware, and is notifying customers. Details: Sakura Internet.
U.S. healthtech firm CareCloud disclosed a breach affecting 3,756,469 individuals after an intruder accessed one of its AWS environments in March 2026 and claimed data exfiltration. Individuals are being offered identity protection via IDX; the firm reported an eight-hour network disruption during the incident. More in CareCloud Breach.
In open-source security coordination, a Linux Foundation–led coalition plans to begin accepting automated, AI-enabled vulnerability reports in September under the Akrites initiative. The program will operate a shared SIRT and a confidentiality-first CVD process, building on CMU’s VINCE platform with LLM-driven deduplication and patch assistance. Background: Akrites Project.