01
Security8 min read
Amazon documents AI agents discovering and exploiting a zero-day in an intrusion
Amazon Threat Intelligence reported that a threat actor used the open-source Cairn agent framework to pursue shell access with little human involvement. Amazon assesses with high confidence that the agents found an unpatched path-traversal flaw in a medical-imaging application and used it for server compromise; it assesses the actor as Chinese-nexus with moderate confidence. The maintainers released a fix after notification.
Why it mattersThe incident compresses vulnerability discovery, exploitation and lateral movement into a machine-speed loop. Defenders should inventory externally reachable AI and medical workloads, deploy behavior-based containment, automate high-confidence response, preserve agent and network traces, and rehearse escalation paths that do not depend on a human reviewing every action first.
Source: AWS Security Blog ↗ · Published November 20, 2026
#Cairn#agentic attacks#zero-day#incident response
02
Security7 min read
AWS STS adds runtime permission scoping for agents and application workers
AWS introduced GetScopedCredentials, an STS API that derives shorter-lived or less-privileged credentials from an existing temporary session. Effective permissions are the intersection of the caller's permissions and an inline scope-down policy, so access can only narrow. The operation can be chained, allowing an orchestrator to issue distinct credentials to sub-agents or tenant-specific workers.
Why it mattersThis supplies a native primitive for replacing ambient cloud authority with task-scoped identity at runtime. Teams should keep policy construction in a trusted broker, bind scope to authenticated task context, isolate the worker from parent credentials, cap duration, log derivation and test that retries cannot broaden access.
Read the updated agent control-plane architecture guide →Source: AWS Security Blog and STS documentation ↗ · Published November 20, 2026
#AWS STS#least privilege#agent identity#temporary credentials
03
AI & Models6 min read
OpenAI adds time awareness and asynchronous user input to the Agents API
OpenAI documented two Agents API capabilities for deadline-bound personal agents. Time awareness periodically reminds the model of the current time, while asynchronous function tools let a run continue useful work while an application waits for user input and incorporate a reply if it arrives. The pattern is intended for agents that must finish by a deadline even when clarification is unavailable.
Why it mattersLong-running agents need an explicit time budget and a resumable wait state rather than a blocking chat turn. Developers should define deadline and fallback behavior, persist checkpoints, make side effects idempotent, expire authority independently of the run and test late, missing and contradictory replies.
Source: OpenAI Developers ↗ · Published November 20, 2026
#Agents API#async input#time awareness#durable workflows
04
Security7 min read
Palo Alto makes model and agent-skill supply-chain scanning generally available
Palo Alto Networks made Prisma AIRS AI Supply Chain Security generally available, combining model and agent-skill inspection in one policy surface. The company says local SDK and CLI scans cover more than 35 model formats and look for load-time code, unsafe serialization, backdoors and malicious skill instructions; proprietary model files remain local while metadata and results go to policy evaluation.
Why it mattersModels and skills can execute code or steer privileged tools, so they need the same immutable promotion controls as packages and container images. Teams should pin digests, retain provenance, scan in an isolated pipeline, validate licenses and sources, require expiring exceptions and keep runtime isolation because a clean scan is evidence rather than proof.
Source: Palo Alto Networks ↗ · Published November 20, 2026
#Prisma AIRS#agent skills#model security#AI supply chain
05
Software Engineering8 min read
PostgreSQL 19 ships concurrent repacking, plan advice and read-your-writes standbys
The PostgreSQL project released version 19. REPACK (CONCURRENTLY) can reclaim or reorder table storage without blocking reads and writes; pg_plan_advice and pg_stash_advice provide plan-stability controls; WAIT FOR LSN enables read-your-writes behavior on standbys; and INSERT ... ON CONFLICT DO SELECT simplifies get-or-create flows. The release also expands logical replication, autovacuum and I/O observability.
Why it mattersSeveral changes reduce operational workarounds in highly available systems, but a major upgrade still changes planner, replication and maintenance behavior. Database teams should benchmark real plans, rehearse logical and physical upgrades, test extensions and failover, verify backup restore and keep rollback criteria before adopting production workloads.
Source: PostgreSQL Global Development Group ↗ · Published November 19, 2026
#PostgreSQL 19#REPACK#logical replication#high availability
06
Security5 min read
Google tells Buildpacks users to rebuild for patched Java runtimes
Google Cloud published patched Buildpacks Java runtime versions 17.0.21.1+2, 21.0.13.1+2 and 25.0.5.1+2 and directed users to rebuild and redeploy applications immediately. Google also says temporary JDK_JAVA_OPTIONS mitigation flags should be removed after workloads are running the updated runtime, referencing security bulletin GCP-2026-082.
Why it mattersSource deployments do not inherit a patched base merely because the platform updated its builder. Teams need rebuild triggers, image and runtime provenance, deployment verification and fleet queries that prove every revision was recreated from the fixed runtime before temporary mitigations are removed.
Source: Google Cloud Buildpacks release notes ↗ · Published November 20, 2026
#Google Buildpacks#Java#runtime security#rebuild
07
Security6 min read
Intel integrates post-quantum acceleration into its cryptography libraries
Intel described instruction-set optimizations for post-quantum algorithms in its cryptography libraries across platforms with AVX2 or newer support. The work targets standardized building blocks including ML-KEM for key establishment and ML-DSA for signatures, positioning software updates—not only future processors—as part of migration planning. Performance statements remain Intel's measurements and require workload validation.
Why it mattersPQC increases key, signature and compute costs across TLS, identity and signing systems. Platform teams should inventory cryptography, benchmark library-supported algorithms on deployed hardware, test hybrid handshakes and certificate growth, preserve interoperability and avoid locking migration to a single processor-specific implementation.
Source: Intel Newsroom ↗ · Published November 20, 2026
#post-quantum cryptography#ML-KEM#ML-DSA#AVX2
08
Security5 min read
NIST decides to revise federal TLS implementation guidance
NIST's Crypto Publication Review Board decided to revise SP 800-52 Revision 2, its guidance for selecting, configuring and using TLS implementations, after reviewing public comments requested in May. NIST has announced the revision decision but has not yet published replacement requirements or a migration schedule.
Why it mattersFederal guidance influences cloud baselines, procurement and regulated security profiles well beyond US agencies. Security architects should track the draft, inventory protocol versions, cipher policy, certificate automation and termination points, but avoid changing production controls before concrete revised requirements are available.
Source: NIST Computer Security Resource Center ↗ · Published November 19, 2026
#NIST#TLS#SP 800-52#security standards
09
Cloud & Infrastructure6 min read
SSC Space and NTT explore an interoperable optical ground-station network
SSC Space and NTT signed a memorandum of understanding to study interoperability between SSC's optical ground-station network and NTT's space terminals, terrestrial optical networks, data centers and network-control technology. The work will examine site diversity, ground-network expansion and service integration; it is an exploratory agreement, not a deployed commercial service.
Why it mattersOptical satellite links promise higher capacity but depend on weather-aware routing and geographically diverse ground sites. Network architects should watch how providers expose link availability, failover, custody, latency and data-center interconnects before treating optical space-to-cloud paths as dependable infrastructure.
Source: SSC Space and NTT ↗ · Published November 20, 2026
#optical communications#ground stations#NTT#satellite networking