
AI Shortens Disclosure Windows as AWS Expands Security Controls
Coverage: 17 Nov 2026 (UTC)
< view all daily briefs >Coordinated disclosure policy and AI-accelerated security operations shaped the news on November 17, 2026. Google's Project Zero and Anthropic shortened their disclosure deadlines in response to AI-assisted vulnerability discovery, while Microsoft made an agent-based security platform generally available. Amazon Web Services released a large batch of updates spanning identity, post-quantum telemetry, compliance automation, agentic AI infrastructure and multicloud networking.
AI Compresses Disclosure and Response Timelines
Google's Project Zero and Anthropic announced substantially faster coordinated disclosure timelines, according to CSO Online. Project Zero will reduce its public disclosure deadline from 90 to 60 days beginning in February, and to 30 days in January 2028. It is also immediately eliminating the previous 30-day soak time that allowed patches to propagate. Anthropic will adopt a 30-day window for high- and critical-severity issues and 45 days for others starting in January. The changes reflect findings that large language models and automated frameworks can rapidly analyze patches, identify underlying bugs and generate exploits, which makes long private reporting windows untenable.
Project Zero also dropped severity as a trigger for timing, because attackers can discover and exploit low-severity flaws as quickly as high-severity ones. It will stop waiting for vendors to verify patches before publishing details. The schedule is intended to push vendors and maintainers to improve remediation processes. However, it increases pressure on open-source projects already overwhelmed with reports, as CVE volume has surged and many teams lack the developer resources to produce reliable fixes. Vendors may lean on AI-assisted patching or alternative mitigations, but studies cited in the report show that AI-generated fixes often miss root causes or provide incomplete coverage. Security teams are advised to prepare for earlier disclosures in three ways:
Harden deployments. Reduce exposure before details become public.
Evaluate compensating controls. Identify mitigations that can be applied while fixes are pending.
Streamline inventories. Make it faster to locate affected systems and apply remediation.
Microsoft described the same shift from the defender's side. The company stated that frontier AI models and accessible open-weight tools have compressed timelines for vulnerability discovery and exploit development from months to minutes. In response, Microsoft announced general availability of Microsoft Perception. The platform orchestrates specialized red, blue and green agents through automated playbooks to simulate attacks, detect and triage alerts, and remediate incidents while keeping human oversight and guardrails in place.
Microsoft also announced several related capabilities:
ISOC in Microsoft Defender. Brings SIEM, UEBA, threat intelligence and built-in Sentinel features to eligible customers.
Codename MDASH. Expanded previews add larger repository scans, regulated cloud support and new MAI-Cyber model variants to improve code scanning and prioritization.
Agent management. Microsoft treats agents as potential insiders and is introducing continuous discovery, onboarding, containment and management capabilities to track agent behavior.
AI Transparency Initiative. Microsoft committed to helping develop open standards for AI transparency, security and governance in collaboration with NIST and more than 40 industry partners.
The post stresses that defenders need to move at machine speed, pair automation with human sign-off, and connect AI usage to cost and outcome metrics.
Identity, Post-Quantum Telemetry and Compliance on AWS
Amazon S3 now records TLS key exchange details in AWS CloudTrail through a new keyExchange field in the tlsDetails section of management and data events. The S3 CloudTrail update is available in all AWS Regions. It shows whether a connection used a post-quantum hybrid algorithm, such as X25519MLKEM768, or a classical exchange, such as x25519. Organizations can measure the share of S3 traffic protected by post-quantum cryptography. They can also identify the SDKs, user agents, applications and IAM principals that still negotiate classical key exchanges and need upgrades. Coverage includes general purpose, directory, table and vector buckets, including traffic via AWS PrivateLink. There is no extra charge beyond standard CloudTrail data event pricing.
Amazon API Gateway REST APIs now support a native JWT authorizer that validates JSON Web Tokens directly in the gateway. This removes the need for Lambda authorizers when using OpenID Connect or OAuth 2.0 identity providers other than Amazon Cognito. Administrators can configure multiple token issuers, accepted audiences per issuer, permitted signing algorithms and additional claims. On every request, the gateway verifies the signature, issuer, audience, expiry and claims. A single authorizer can be attached to multiple methods while enforcing different scope requirements per method. The feature covers Regional, edge-optimized and private REST APIs in all commercial AWS Regions and AWS GovCloud (US).
Amazon ECR introduced global authentication tokens for private registries. A single token requested via the AWS Sign-In CreateBearerToken API works across all ECR private registries within an AWS partition. Previously, separate tokens were required per Region for push and pull operations. This simplifies proxy services that route container pulls across Regions. Global tokens have a default lifetime of 12 hours, configurable down to five minutes to meet shorter-lived credential policies. They work with the Docker CLI and the Amazon ECR Docker Credential Helper, and existing regional tokens continue to function as before.
AWS Config added managed conformance packs. AWS publishes and maintains definitions for frameworks including PCI DSS v4, SOC 2, NIST SP 800-53 Rev. 5, ISO/IEC 27001:2022 and CIS Controls. AI governance frameworks are also covered, including the EU AI Act and NIST AI RMF 1.0. Deployed packs update automatically when new managed rules become available in a Region or when a framework is revised. Definitions are uniform across supported commercial Regions, excluding UAE and Bahrain at launch. Customers can still set rule parameters, and evaluations are billed at standard conformance pack rates.
Securing and Building Agentic AI Workloads
An AWS Security Blog post on exposing AgentCore Runtime addresses the ingress path. AgentCore Runtime isolates each session in a dedicated microVM, and AgentCore Identity issues workload-specific identities. Even so, the ingress path remains the customer's responsibility under the AWS Shared Responsibility Model. The guidance centers on identity propagation: short-lived, user-scoped tokens are passed through each hop so that each component can only call the next one, which limits the blast radius if a segment is compromised. The post presents two patterns:
Pattern A. Amazon API Gateway with a Lambda authorizer and a private Lambda proxy that forwards the user token through a VPC endpoint. It adds DynamoDB-based per-user throttling, AWS WAF, Guardrails and a CloudWatch audit trail.
Pattern B. The managed AgentCore Gateway centralizes ingress authentication, token exchange, routing, native rate limiting and WAF protections, without custom proxies or VPC networking.
The post recommends defense-in-depth controls to prevent credential abuse, sandbox escapes and cross-tenant data leakage. These include authentication at multiple layers, session binding, per-user throttles, request inspection and continuous token validation.
Separately, Bedrock AgentCore now includes free web search queries, with an allowance of 5,000 per month through the Web Search Tool in the AgentCore gateway and the Web Search API. The allowance applies automatically, is shared across supported Regions and all accounts in an organization, and resets monthly. Queries beyond the free tier are billed at standard rates, and billing does not interrupt running agents.
AWS also made OpenAI GPT-6.2 models Astra and Sol generally available on Amazon Bedrock. Astra targets high-stakes tasks that require advanced reasoning, such as investigating complex codebases, and includes an Ultrafast mode for latency-sensitive scenarios. Sol targets routine, high-volume professional and coding workflows. AWS states that established controls are available to secure workloads, govern access and audit model invocations.
Amazon SageMaker launched AI data preparation capabilities, which include:
Synthetic data generation grounded in a customer's own documents.
Auto-labeling with confidence scores, with low-confidence rows routed for human review in Label Studio Community Edition within SageMaker AI Studio.
Data quality evaluation and transformation operators for cleaning and reshaping datasets.
Source data, generated datasets and labels remain in customer-controlled S3 buckets governed by the customer's IAM roles. The features are initially available in US East (N. Virginia) and US West (Oregon).
Amazon Elastic VMware Service added support for r7i.metal instances for VMware Cloud Foundation 9.x. Because these instances lack local NVMe disks, EVS now supports external principal storage such as Amazon FSx for NetApp ONTAP. AWS also released an Agent Skill that lets AI agents plan, deploy and operate EVS environments, including driving deployments through the EVS Deployment Orchestrator.
Multicloud Connectivity, Network Resilience and Compute
Google Cloud and Microsoft introduced Partner Cross-Cloud Interconnect for Microsoft Azure in preview. The service provides programmatic, private, high-bandwidth connections between the two clouds in minutes, without ordering physical circuits. Key details:
Support and redundancy. The service is jointly supported by both vendors and includes redundancy across separate physical locations.
Encryption. Traffic between cloud edge routers is encrypted automatically.
Pricing. A fixed rate applies for the chosen bandwidth.
Availability. Initially offered in us-east4, us-west2 and europe-west3.
Google says the service is built on an open specification for network interoperability, a collaboration that began with AWS.
On the same day, AWS announced general availability of AWS Interconnect - multicloud with Microsoft Azure, following an Azure public preview in August. Customers can now interconnect workloads across AWS, Azure, Google Cloud and Oracle Cloud Infrastructure through a consistent interface. The initial Azure-enabled Regions are us-east-1, ap-southeast-2 and eu-central-1.
Amazon VPC, AWS Transit Gateway and AWS Cloud WAN now offer a 99.999% availability SLA for cross-Region traffic. The SLA is enabled by a network-level feature called Multipath Resilience, which duplicates packets across physically diverse fiber paths on the AWS backbone and delivers whichever copy arrives first. The feature targets payment processing, financial trading, telephony and live media streaming, which cannot tolerate interruptions from fiber cuts or partial network degradation. It is generally available in US East (N. Virginia), US East (Ohio), US West (N. California), US West (Oregon) and Europe (Ireland).
Compute updates focused on network-intensive workloads. EC2 C8ine instances are now available in Europe (Ireland). They use custom sixth-generation Intel Xeon Scalable processors and sixth-generation AWS Nitro cards. Amazon cites up to 43% higher compute performance and up to 2.5x higher packet performance per vCPU compared with C6in, positioning the instances for virtual firewalls, load balancers and Telco 5G UPF deployments. Graviton4-based EC2 C8gn instances reached Israel (Tel Aviv). They offer up to 30% better compute performance than C7gn and up to 600 Gbps networking.
Amazon MSK Express Brokers now support Graviton4 M8g instances. These deliver 15% lower compute costs than M7g and up to 45% lower P99 produce latency. AWS also removed the broker-side charge for cross-AZ data transfer between clients and brokers. Separately, Lambda Managed Instances can now scale to and from zero for Event Functions when MinExecutionEnvironments is set to zero.
DynamoDB caching is now generally available with Amazon ElastiCache Serverless for Valkey. It provides a write-through cache for sub-millisecond reads without application code changes. Access control remains consistent because applications use their existing IAM credentials, and AWS does not charge extra for the caching feature itself.