When Trusted Access Turns Against You

In the News

DevOps Related News

OpenAI’s September 29 DevDay gets a longer look this episode; the announcements below describe commercial offerings, with government-cloud availability for these new features still unconfirmed.

• OpenAI DevDay — GPT-6.1 Sol and Astra Ultrafast: GPT-6.1 Sol brings lower-cost coding and professional work than Astra, with multi-agent delegation in beta; Astra Ultrafast speeds up output in the API, Codex, and ChatGPT Work. Ultrafast’s API supports U.S. data residency but not other inference regions, while app access requires the $500 Pro plan or eligible Enterprise/Edu plans. API release notes · DevDay roundup

• OpenAI DevDay — cloud coding and browser agents: Codex can prepare and test a reusable cloud environment, then start each task in an isolated workspace that keeps running while your computer sleeps. Separately, the Agents API gained an OpenAI-hosted browser, with website approvals and sign-in handled by the application using it. Cloud environments · API release notes

• OpenAI DevDay — dots, synced work, and Space: Dots are always-on agents that keep working between conversations and bring decisions back to you, with access rolling out gradually. New synced Work tasks can continue across desktop, web, and mobile, while Space brings shared files and editable Pages into one place; steps that need your local computer still require it to be online. Dots · DevDay roundup

• OpenAI DevDay — sign-in and connected Sites: Sign in with ChatGPT lets participating apps use a ChatGPT identity and, with separate permission, eligible users’ plan usage; commercial sign-in remains a limited partner trial. Workspace-private Sites can also use each visitor’s connected apps where enabled, with that visitor choosing the account and consenting to access. Sign-in documentation · Rollout details

• OpenAI DevDay — plugin interfaces and website feedback: Plugin Extensions add sidebar apps, conversation panels, file viewers, editors, and forms inside ChatGPT; composer mentions are desktop-only and Free/Go web extensions are still coming. The Browser Annotation API lets developers attach context and preview controls to selected website content in supported desktop browsers. Plugin Extensions · DevDay roundup

• OpenAI DevDay — event-driven teamwork: MCP Events lets users subscribe to server updates through webhooks, using MCP 2.0 and a still-draft event specification. Team Tasks run scheduled or event-triggered work through team service accounts and company-managed workspace connections, while Slack and Teams deployments let administrators choose who can use them and which channels they can access. MCP Events · Team Tasks · DevDay roundup

• Git 2.56: The September 28 release coverage highlights new commands for dropping commits and managing references, plus a way to inspect remote object metadata without downloading the full object. That last change saves bandwidth when a partial clone only needs to know a file’s size, not fetch its contents. Git contributor walkthrough

• GitLab 19.4: Agents can now start, retry, cancel, and inspect CI pipelines through GitLab’s MCP tools, including on Self-Managed installations. There’s an upgrade catch for Geo users: SSH proxying is now on by default, so cloud-native deployments using bundled NGINX Ingress must move to Envoy Gateway or disable the two proxy flags to prevent stalled SSH operations. Release notes

• Kubernetes 1.37: September’s feature walkthroughs cover new alpha controls for emptyDir permissions and bind mounts, alongside Memory QoS reaching beta on cgroup v2 Linux nodes. Memory QoS is enabled by default, but it won’t start reserving or throttling memory until those settings are configured. Storage hardening · Memory QoS

• Managed Kubernetes — GKE Agent Substrate: Google’s new open-source runtime puts agents in isolated sandboxes with controlled network access, and it can run on Kubernetes outside GKE. On GKE, non-production access is open and production access requires an allowlist; government-cloud availability hasn’t been confirmed. Google Cloud announcement

• Keycloak 26.7.4: The self-hosted identity platform fixes slowdowns, Oracle database startup failures, and errors when administrators manage subgroups. The September 16 release also tightens replay protection and admin permissions. Keycloak release notes

• Sonatype Nexus Repository 3.96.3: The September 22 self-hosted release fixes a storage-migration bug that could delete artifacts belonging to other repositories sharing the same blob store. It follows September 18’s NuGet fixes for failed package uploads and missing debugging-symbol metadata. Sonatype release notes

Cyber Security

• OpenAI DevDay — Codex Security Cloud: The research-preview plugin scans connected GitHub repositories and can monitor new commits, with findings, validation evidence, and proposed patches available for review before a draft pull request is created. This is a hosted workflow; the announcement does not establish GitLab support or government authorization for the new service. Security Cloud documentation · DevDay release status

• OpenAI DevDay — Agent Security controls: The new admin surface brings agent permissions and execution settings under a shared policy baseline, with separate settings for local execution and Codex Cloud. Work Cloud retains its own execution controls, so a restriction on a developer’s laptop does not automatically protect work running in a cloud container. Agent Security documentation

• Storm-3168’s destructive Azure attack: Microsoft’s September 25 report details how compromised application identities were used in a June intrusion to attempt more than 100 storage-account deletions in seven minutes, with most succeeding. Some deletion safeguards held, but a Key Vault and application resources were also deleted; Microsoft did not confirm data theft or observe a ransom note. Microsoft’s investigation

• Brevo JavaScript supply-chain attack: Sansec’s September 16 investigation found that Brevo’s embedded scripts delivered malicious code on September 14, with the firm reporting exposure across more than 100,000 customer sites—not 100,000 confirmed site takeovers. The code tried to install a malicious plugin through logged-in WordPress administrators and showed other visitors a fake verification prompt that tricked them into running a command. Sansec’s technical investigation

• EvilTokens account-compromise campaign: Microsoft announced a September 22 takedown of a cybercrime service linked to more than 12,000 compromised inboxes across over 10,000 organizations. Its AI assistant searched stolen mail for payment discussions and trusted contacts, then helped attackers plan fraud and impersonate the right people. Microsoft’s takedown report

• Star Blizzard’s expanded espionage campaign: Microsoft’s September 29 report describes Russian state-linked phishing activity affecting more than 100 organizations, primarily in the U.S. and U.K., including government and policy-related targets. The group’s RedFlick technique uses disguised files and scheduled tasks to install a backdoor with fewer steps from the victim than its earlier fake-verification attacks. Microsoft’s campaign analysis

• Google’s agentic code security: Google shared how its internal Mantis workflow reviews code changes for security problems and checks whether the suspected vulnerable code can actually be reached. A separate agent drafts fixes, with engineers still reviewing the patches before they go in. Google engineering report

• Sysdig CLI Scanner 1.30.0: Sysdig can now find vulnerable binaries that were copied into an image without a package manager, check binary allow/deny lists, and report package licenses in technical preview. Offline scans can use a local vulnerability database, but don’t run the policies managed by the backend. Release notes · Offline scanning

Compliance and Government

• Harvester Government STIG released: DISA approved the Rancher Government Solutions Harvester Government Security Technical Implementation Guide, announced September 18 and effective upon release. That gives the government edition an official hardening baseline, not an automatic authorization to operate for every deployment. DISA announcement · Official STIG library
• NIST’s DevSecOps implementation guidance: NIST’s September 24 update adds a second example implementation covering CI/CD automation and container deployment, along with mappings to the Secure Software Development Framework. The material is open for comment through November 9, while a planned third implementation will explore how to identify, authenticate, and authorize AI agents in the software lifecycle. NIST project update
• CISA vulnerability-bulletin retirement: CISA set September 28 as the cutoff for its weekly vulnerability bulletin, shifting emphasis from severity scores to which flaws pose real risk. Its September 16 notice directs subscribers to the exploited-vulnerability catalog, security advisories, and vendor alerts instead. CISA’s notice
• NGA search-platform acquisition: NGA wants to bring its data catalog and search tools together so users can find information across separate data stores, with cloud access and automated workflows built into the plan. Its September 17 notice points to an industry information request in the first quarter of fiscal 2027, followed by a formal acquisition later that year. Intelligence Community News · NGA notice
• NITAAC’s IT-contract transition: NITAAC used a September 16 interview to explain why it canceled CIO-SP4 back in February: too much overlap with contracts already available through GSA and NASA. The deadline still ahead is October 29, when new ordering ends on CIO-SP3, CIO-SP3 Small Business, and CIO-CS. Federal News Network interview · NITAAC transition notice
• FedRAMP deadline guidance: FedRAMP updated its consolidated-rules guidance on September 22 to explain which dates apply to getting certified, maintaining certification, optional adoption, and grace periods. This clarifies how to read the Rev5 and 20x deadlines; it doesn’t introduce a new set of deadlines. Official changelog

Industry News

• Everfox deployment at the FBI: The FBI announced that Everfox’s insider-threat monitoring is now operational across multiple bureau networks and tens of thousands of endpoints. The September 23 announcement describes user-activity monitoring and behavioral analysis deployed inside government networks, not just a product being offered to federal buyers. FBI announcement
• Vectra AI government authorization: Vectra AI and Carahsoft announced FedRAMP High authorization for a platform that connects threat signals across networks, identities, cloud services, and SaaS. Federal buyers can purchase it through contracts including GSA and NASA SEWP. September 15 announcement
• Commvault government recovery services: Qualified U.S. managed-service providers can now deliver Commvault’s FedRAMP High-authorized Cloud for Government. The change gives agencies another way to buy managed backup and recovery for workloads spread across on-premises systems, cloud infrastructure, and SaaS. September 15 announcement
• Owl cross-domain data transfer: Owl announced government cross-domain certification for V2CDS 2.1, which adds support for checking structured data as it moves between networks with different security classifications. It runs on enterprise servers and rugged edge hardware, though agencies still need their own authorization to operate. Owl’s September 17 release
• Parsons defense software procurement: Nine Parsons offerings earned awardable status across the Tradewinds and Platform One marketplaces, including the Kubernetes-based METAL data platform and TReX testing tool. That clears a purchasing hurdle for government buyers, but doesn’t mean nine contracts have been awarded. Parsons’ September 14 announcement
• Claude Opus 5.5: Anthropic released Opus 5.5 on September 22 for longer coding and knowledge-work tasks, reporting lower running costs than Opus 5. AWS also announced availability in GovCloud through Bedrock, with zero-data-retention support by default. Anthropic release · AWS GovCloud announcement

Community News

• Jev and smaller AI decisions: Unicodeveloper’s guide explores using Jev to classify and route work while ordinary code stays in charge of the workflow, building on TypeSafe’s September 15 launch. It’s a useful design discussion, but a correctly formatted answer can still be wrong, and Jev’s government or self-hosted availability remains unconfirmed. Community guide · TypeSafe’s explanation · Open-source comparison adapter
• Trendyol’s SRE AI agents — August 24 read: Trendyol describes an in-house incident investigator that gathers metrics, traces, Kubernetes state, and automated checks in parallel, then cross-checks the evidence before proposing a root cause. Its Kubernetes specialist has read-only access, and SREs can question the report after the automated run. Engineering walkthrough

Fun Tools and Reads

• Deleting 3,000 lines of AI-generated code — September 9 exception: Devrim Ozcay’s essay asks a pointed question: are passing tests enough if the engineer responsible doesn’t feel able to troubleshoot the feature in the middle of the night? Read the essay
• Container security checklist: Neel Shah’s September 21 guide walks through non-root containers, read-only filesystems, minimal images, restricted Linux privileges, and Kubernetes admission policies that enforce those settings. Read the guide
• AI-assisted EC2 right-sizing — August read: This walkthrough combines 30 days of CloudWatch metrics with the Claude API to draft instance-size recommendations, while flagging that EC2 memory use needs CloudWatch Agent data before a downsize decision. Read the walkthrough