More automation is arriving with more decisions attached: who can send data outside Workspace, who reviews an agent's finished work, and what happens when a build runner changes underneath you. Meanwhile, Brevo's incident shows why a clean origin server can miss an attack, and Excel's paste fix comes with an exception worth putting in the helpdesk note.
What matters most today
Brevo has removed the malicious injection, but affected visitors and WordPress sites may still need investigation. A clean upstream script does not establish a clean downstream system.
Claude coordinates parallel work; Junie records UI checks. Faster execution helps only when reviewers can see the result, resolve conflicts, and distinguish a completed run from a passed test.
GitHub's runner migration has dates. Workspace Studio's new capabilities have approval and edition limits. Use the lead time to make a deliberate rollout decision.
Excel 2016's MSI paste fix retains a conditional-formatting caveat. Give support staff the affected edition and remaining workaround, not a blanket all-clear.
AWS can preserve SFTP client addresses behind a load balancer. netlab is changing its preferred virtualization path. Both warrant a representative test before changing established operations.
Action / Watch List
-
Investigate
If your site embedded Brevo forms, Conversations, or SDK scripts during 2026-09-14 16:07–20:30 UTC, review related visitor reports and WordPress plugin changes. Escalate unexpected plugin installations or reports of a pasted verification command to incident response; removal of Brevo's injection does not remove resulting malware.
-
Patch
If you run Cisco ISE or ISE-PIC, install your branch's fixed release: 3.1 Patch 12, 3.2 Patch 11, 3.3 Patch 12, 3.4 Patch 7, or 3.5 Patch 4. Review every node and external network logs using Cisco's compromise guidance; suspected compromise requires recovery as well as patching.
-
Restore
For Excel 2016 MSI installations affected by KB5002914, apply KB5002665 and try the previously failing paste on a workbook copy. If conditional formatting still causes failure, give users Microsoft's documented Paste Special options; do not apply this MSI package to Click-to-Run installations.
-
Pilot
If you use Anthropic's Claude Code and the redesigned Projects beta becomes available on your Claude Pro or Claude Max plan, try it on a small coding task. Check the proposed code changes, test results, conflicts between changes, and subscription usage to see whether it saves your team time overall.
-
Test
If your team uses JetBrains' Junie coding agent, try its CLI /demo command to check one application interface change. Configure a Docker demo environment and specify one visible action and its expected result. Review the recording and pass/fail verdict before accepting the code change; verify that the demonstration actually exercised the intended behavior.
-
Compare
If your n8n Cloud plan offers AI Gateway credits, test an AI workflow with non-sensitive data. Compare the cost of paying for model requests through n8n's shared credit balance with using your own model-provider API keys. Confirm your organization approves the selected provider to receive the workflow's data before using Gateway credits in production.
-
Prepare
If you administer Google Workspace Studio, Google's workflow automation tool, decide which actions need human approval before enabling its new connections to external apps or its webhooks, which send requests to external services. To restrict which web addresses those webhooks can contact, configure the webhook URL allowlist. That control requires Google Workspace Business Plus, Enterprise Standard or Plus, or Education Standard or Plus. If your organization requires destination restrictions and your edition lacks that control, leave Workspace Studio webhooks disabled.
-
Test
If your GitHub Actions workflows use GitHub-hosted runners labeled ubuntu-latest, test representative builds with the ubuntu-26.04 runner image before 2026-10-19. Resolve missing tools and version changes, or explicitly select ubuntu-24.04 while you prepare; GitHub's automatic migration of ubuntu-latest runs through 2026-11-19.
-
Test
If your AWS Transfer Family SFTP server uses a VPC endpoint behind your own Network Load Balancer, test the new source IP preservation option on a test server. Confirm the server logs and your custom identity provider receive the client's original IP address and that allowed clients still authenticate before changing production access rules.
-
Migrate
If network validation depends on netlab's Vagrant/libvirt provider, reproduce one important topology with containerlab and suitable vrnetlab images. Compare device behavior and automation results before migrating more labs; retain the existing setup where device support is incomplete.
AI / Agents / Developer Workflow
Claude Code projects become a coordinator for parallel cloud work
Brief: Anthropic introduced redesigned projects in a limited Claude Code beta. A coordinator scopes work, delegates to parallel cloud threads, and assembles results; initial access is for selected Pro and Max users without existing web or desktop projects.
Operational Impact: Pilot a bounded repository task before relying on it for delivery. Threads use separate branches and repository copies, so overlapping changes still produce merge conflicts. Multiple full sessions can consume usage faster, and Team and Enterprise availability follows later. Shared project memory also makes recorded decisions reusable across threads, so review the context as well as the code.
Strategic Context: The tool is taking on coordination as well as implementation. The useful comparison is total time to an accepted change, including human review and integration, rather than the number of threads running at once.
Junie /demo gives reviewers a recording of the changed behavior
Brief: JetBrains introduced /demo in Junie CLI to build an application, interact with its UI, and produce video, screenshots, and an HTML report. It uses a Docker environment and a model with computer-use support.
Operational Impact: Start with a visible feature and an explicit expected result, then inspect what the agent actually exercised. JetBrains' CI example treats FAIL, PARTIAL, and missing or invalid verdicts as failures. Budget for model use and the CI runner separately before expanding coverage.
Strategic Context: Generated code creates more work for reviewers unless evidence travels with it. A recording can help show behavior, but its value depends on the scenario exercised; it does not establish that untested paths are correct.
n8n Gateway credits remove account setup from supported service trials
Brief: n8n announced Gateway credits for supported AI models and tools on select Cloud plans, starting with v2.36. A shared prepaid balance replaces separate provider credentials for six model providers and five tool services.
Operational Impact: This can shorten a trial, but the selected provider still receives request content. Instance owners can disable Gateway credits, and the Cloud Admin Panel breaks usage down by service and workflow. Compare actual charges and data-handling requirements before moving an established automation. The balance is shared across the instance, with automatic top-ups available; establish who can replenish it and review workflow spending.
Strategic Context: Consolidated billing removes some administrative work without removing the underlying suppliers. The buying decision remains about cost, approved data destinations, and whether keeping direct provider credentials gives the team useful control.
IT Ops / Security / Infrastructure
Brevo's ClickFix post-mortem points to downstream cleanup
Brief: Brevo's newly published post-mortem says a stolen Cloudflare API key enabled malicious edge responses on 2026-09-14. Embedded customer scripts were affected from 16:07 to 20:30 UTC, including attempts to install a plugin when a WordPress administrator visited.
Operational Impact: Site owners should investigate that exposure window and any visitor who followed the fake verification instructions. Brevo removed the Worker and revoked credentials, but those steps do not establish whether a visitor device or customer site remains compromised. Escalate suspicious plugin changes for incident response.
Strategic Context: The origin files were unchanged while responses were modified at the CDN edge. Monitoring only the files on the server leaves a gap in what customers actually receive from third-party scripts.
Cisco ISE remains an urgent patch-and-investigate item
Brief: Cisco confirms active exploitation of CVE-2026-76460, an unauthenticated API bypass affecting ISE and ISE-PIC regardless of configuration. Its advisory provides fixed branches and warns that attackers may obtain root command execution.
Operational Impact: Complete the appropriate branch upgrade and review access logs on every node. Check external network and firewall logs too, because an attacker can hide evidence on the appliance. Cisco recommends re-imaging affected nodes and restoring configuration as needed when malicious activity is suspected.
Strategic Context: This is a second-day carry-forward because exploitation keeps the work urgent. Installing the fix closes the vulnerability; establishing whether an identity appliance was already compromised is a separate recovery decision.
Platforms / Devices / Buying Signals
Workspace Studio gains external automation with controls to set first
Brief: Google announced custom starters and steps, third-party integrations in beta, and outbound webhooks for Workspace Studio. Admin settings began rolling out on 2026-09-17; end-user rollout starts on 2026-09-21 for Rapid Release and 2026-09-30 for Scheduled Release.
Operational Impact: The capabilities are off by default. Choose approval requirements before enabling them, and check edition coverage: webhook URL allowlists require Business Plus, Enterprise Standard or Plus, or Education Standard or Plus. A small pilot should establish which external actions need human approval. Custom steps and integrations have separate approval settings, while webhooks follow Sensitive Steps approvals; one setting does not govern everything.
Strategic Context: Office automation is reaching outside the suite. Ease of connection matters less than whether the administrator can restrict destinations and keep approvals aligned with the business action being performed.
GitHub Actions sets the Ubuntu 26.04 migration window
Brief: Ubuntu 26.04 GitHub Actions runners are generally available for x64 and arm64. The ubuntu-latest label will move from Ubuntu 24.04 to 26.04 gradually between 2026-10-19 and 2026-11-19.
Operational Impact: Run representative builds against ubuntu-26.04 now and identify dependencies on preinstalled tools or versions that changed. If a workflow is not ready, GitHub recommends pinning ubuntu-24.04. Give that temporary pin an owner so it does not become an undocumented permanent dependency.
Strategic Context: A floating runner label buys maintenance convenience at the cost of a moving environment. The announcement provides time to turn an automatic platform change into a controlled build migration.
AWS Transfer Family can retain the real SFTP client address
Brief: AWS Transfer Family added source IP preservation for SFTP servers with VPC-hosted endpoints behind a customer's Network Load Balancer. Proxy Protocol v2 carries the client address into logs, events, and custom identity-provider authentication.
Operational Impact: If existing records show the load balancer's private address, test this per-server setting before changing access rules. Check the address reaching your identity provider and representative allowed-client logins. The feature is available in all Regions where Transfer Family is offered.
Strategic Context: A network intermediary can erase information that an application depends on for auditing or authorization. Preserving that information makes those controls more useful, but changing the address seen by policy also deserves rollout testing.
User-Facing Apps / Platform Friction
Excel 2016 gets a paste fix with a conditional-formatting caveat
Brief: Microsoft's KB5002665 addresses silent paste failures introduced by KB5002914 in Excel 2016. The package applies to MSI installations, not Click-to-Run, and Microsoft says pasting may still fail in workbooks containing conditional formatting.
Operational Impact: Match the installation type before deploying the fix, then test a copy of an affected workbook. For remaining failures, Microsoft documents Paste Special options including formulas, values, and links. Support guidance should explain that the patch does not establish a universal resolution across Excel editions or workbook types.
Strategic Context: Silent failures are particularly disruptive in spreadsheets because the user may assume the data moved. This unresolved support problem earns space despite the earlier patch date; the important distinction is the tested workflow, not just an installed update.
Infrastructure / Self-Hosting
netlab winds down Vagrant/libvirt development without an immediate removal
Brief: netlab's maintainer says the project will rely primarily on containerlab and stop adding libvirt features or running device integration tests with that provider. Existing code remains, bug fixes continue, and platform integration tests remain for now.
Operational Impact: Network teams using these labs for change validation should trial important device images through vrnetlab and compare a representative topology. Keep working Vagrant environments available while confirming coverage. The announcement does not require an immediate teardown or supply a firm code-removal date. Continued platform integration tests can catch general breakage, but they do not replace the device-specific coverage the maintainer is discontinuing.
Strategic Context: Test infrastructure has a support lifecycle too. A lab that still starts can gradually lose assurance as device testing moves elsewhere; migrating the validation path matters before that divergence reaches a production network change.
Coverage notes
Exact scan window: 2026-09-17 09:00:58 MDT through 2026-09-18 08:56:41 MDT, America/Denver. The last completed retained digest cutoff was 2026-09-17 09:00:58 MDT, taken from its source metadata. This is not a first-run fallback.
Ten full cards use ten distinct primary sources. The private allocation is three AI/developer/automation, three platform/enterprise, two security-action, and two user-facing/infrastructure cards, with no double-counting. Security is below its usual target; older advisory recaps were not added to fill the lane. Careers and policy searches did not yield an additional sufficiently verified current item for a full card.
Three cards use explicit freshness exceptions: Cisco's 2026-09-16 advisory remains urgent because of active exploitation and is repeated for only the second consecutive day; Excel's 2026-09-16 fix has unresolved support impact; netlab's 2026-09-17 announcement predates the scan start but was missed in the prior digest and changes current lab-support planning. No older item was added solely for balance.
Brevo's incident occurred on 2026-09-14; its post-mortem was reported as newly published on 2026-09-17. The full card uses that publication date and the directly inspected vendor account, including the narrower embedded-script exposure window. The card does not imply the upstream injection is still active.
Official release notes, product announcements, Microsoft support guidance, the Cisco advisory, Brevo's post-mortem, and netlab's maintainer statement were inspected directly. CISA's current KEV page and JSON feed could not be retrieved; no new catalog addition or federal remediation deadline is asserted. Cisco's alternate official support page was readable after its security-center URL failed.
The miss-check covered AI releases, coding agents, Microsoft identity and app issues, security advisories, outages, pricing, policy, and workforce. Current search activity around GPT-Live-1 pointed to a 2026-09-10 launch and was excluded as stale. Aggregator and forum signals were used only as leads; no full card relies on rumor. No Radar assignment or editorial-context file was available locally for this run.