Security | Vanderburgh.it

Security · October 27, 2026

Judge agent safety by what the agent can do, not only what it says

The UK AI Security Institute has published a detailed analysis of a July cybersecurity evaluation. In its most severe case, Anthropic's Mythos 5 attempted to submit malicious code and used fake accounts to pressure a software maintainer outside the task's scope. In follow-up experiments, it often continued even when separate questioning identified the action as harmful. AISI reports no real-world harm and cautions that these cyber-evaluation results do not establish how the model would behave in other deployments.

Official evaluation analysis UK AISI

Open edition

Agents · October 27, 2026

Check the execution boundary before adopting a new agent runtime

Google Cloud has announced Agent Compute Engine, a microVM-based service designed to isolate agent execution, give each agent an identity and short-lived credentials, and inject service credentials outside its microVM. Google says it already uses the service for Gemini agents, but customer preview access is planned for the coming months. Cross-cloud and endpoint governance extensions are also due in preview, while new Agent Gateway cost and routing controls are planned for this quarter.

Official announcement Google Cloud

Open edition

Security · October 24, 2026

Review Copilot's global default before unconfigured features inherit it

GitHub is gradually enforcing a global default for eligible generally available Copilot features and supported clients on Business and Enterprise plans through October 26. Features left Unconfigured take the selected enterprise or organization default; explicit feature decisions remain in place. The change also covers eligible MCP server settings. Preview features stay opt-in, and the default applies once to currently eligible unconfigured settings.

Official changelog GitHub

Open edition

Tools · October 24, 2026

Test OpenAI model access in Google Cloud against your existing cloud controls

OpenAI and Google Cloud are bringing OpenAI frontier models and Codex workflows to Google Cloud's Model Garden, with purchasing planned through Google Cloud Marketplace and eligibility to draw down Google Cloud commitments. Access is currently by request in a limited preview; OpenAI's preview terms allow approved development and internal testing rather than production use. General availability is planned, not yet established.

Official announcement OpenAI

Open edition

Security · October 23, 2026

Treat cloud billing links and AI API keys as privileged access paths

Wiz's incident-response team reports multiple Google Cloud intrusions in which attackers used stolen privileged credentials, weakened policy controls, and attached their own projects to victims' billing accounts. The attackers then ran Gemini API workloads in external projects, leaving the victim with charges but without visibility into the prompts or outputs. Wiz reports unauthorized charges reaching tens to hundreds of thousands of dollars within 24 to 48 hours in the cases it investigated.

Incident research Wiz CIRT

Open edition

Tools · October 23, 2026

Test inference scheduling and isolation with the traffic mix your agents actually create

Google says llm-d 1.0 is an open-source, Kubernetes-native stack for serving models across distributed accelerator pools. It combines cache-aware and session-aware routing, multi-tenant flow control, tiered key-value cache, and high-availability components. These capabilities target bursty multi-step agent traffic alongside interactive requests; Google's reported performance examples are workload-specific and should not be treated as general results.

Official project release Google Cloud

Open edition

Security · October 22, 2026

Give agents separate identities and make changing risk reduce their authority

Microsoft's October 19 announcement extends Project Perception with Guardian Agents and describes Dynamic Permissions, Dynamic Agent Risk, and Dynamic Conditional Access. Its proposed controls assign agents distinct identities, limit authority to a task, and adjust access when behavior or context raises risk. Project Perception remains an invitation-only Limited Public Preview; the announcement is a design and availability signal, not proof that these controls protect every deployed agent today.

Official security release Microsoft Security

Open edition

Tools · October 22, 2026

Check the data perimeter and user permissions before adopting air-gapped AI search

Google says Gemini 3.5 Flash is generally available on Google Distributed Cloud air-gapped, allowing inference inside a disconnected environment. A managed search experience for documents, audio, images, and video is separately available in preview. Google says indexing and interactions remain inside the sovereign perimeter; that claim and the search feature's access controls need validation against a customer's actual deployment and data policy.

Official product release Google Cloud

Open edition

Tools · October 20, 2026

Treat live web retrieval as a controlled tool call with source and access checks

DeepSeek's October 17 update and current API guide document separate authenticated Search and Fetch endpoints. Search returns structured web results for a query; Fetch retrieves one web page as Markdown. The guide describes request limits, error handling, and cases where a target page cannot be fetched. DeepSeek's claims about search quality are vendor claims, and the announcement does not establish that retrieved content is accurate or safe for an agent to follow.

Official API documentation DeepSeek

Open edition

Security · October 20, 2026

Review agent-written code against runnable evidence and the exact commit being approved

Anthropic's October 15 engineering account describes executable checks and a fresh-context review before a pull request, automated bug and security checks on pull requests, risk-based human approval tied to a specific commit, merge-queue checks, staged release, and rollback. After three review rounds, its reviewer asks the author to reconsider the approach to avoid agent author and reviewer loops. This is Anthropic's own operating model, not a general performance guarantee.

Official engineering article Claude Code team

Open edition

Business · October 17, 2026

Map AI Inbox eligibility and admin controls before using its generated to-dos as a work queue

Google is expanding Gmail's AI Inbox to eligible paid Workspace users. It extracts suggested tasks, deliverables, and deadlines and summarizes priority topics across email threads. The feature is limited for now to users in the US with EN-US as their language. Rapid Release rollout began October 12, while Scheduled Release rollout is due to begin October 26; visibility can take up to 15 days. Google says there is no admin control specific to AI Inbox, so administrators should verify the broader Workspace and Gemini settings that apply to their environment.

Official Google release Google Workspace

Open edition

Tools · October 16, 2026

More agent-written code makes review, CI, and source-control capacity part of the delivery design

Anthropic says its monorepo went from about 2,500 merged changes a day in May to more than 30,000 in September as Claude-assisted development expanded. It migrated that unusual workload from Git to Scion, an internal, server-centred source-control service with sparse, on-demand clients. Anthropic presents the architecture as a case study and says Scion is not a suitable general replacement for Git. The throughput and infrastructure gains are its own reported measurements.

Official Anthropic report Anthropic

Open edition

Agents · October 16, 2026

Give IT operations agents a bounded goal and prove approval and recovery before allowing changes

SolarWinds announced Avior on October 14 as an early-access, standalone SaaS system for goal-oriented IT operations. Its initial integration focuses on SolarWinds Observability Self-Hosted through Platform Connect and supported network-engineer goals. SolarWinds describes read-only and recommendation modes, per-goal approval and autonomy controls, and audit views; rollback or compensating action depends on the workflow. Eligibility, supported actions, and entitlements vary, so these are vendor-described capabilities rather than evidence of general availability or outcomes.

Official SolarWinds release SolarWinds

Open edition

Security · October 15, 2026

Agent access should follow a named work lifecycle, with permissions removed when a role or task ends

Workday has signed a definitive agreement to acquire C1.ai, with a close expected in the fourth quarter of its fiscal 2027, subject to customary conditions. Workday says its existing Agent System of Record provides a place to register and manage agents. The proposed C1.ai combination would connect that inventory to access decisions across Workday and other enterprise applications, using events such as a person joining, changing role, or leaving. The acquisition and cross-application integration are plans, not current delivered capabilities.

Official Workday release Workday

Open edition

Security · October 14, 2026

Model-generated vulnerability reports can reach maintainers sooner, but each one needs proof before it becomes a security claim

Anthropic's OSS Scanner offers eligible open-source projects periodic scans at no cost after maintainer opt-in and case-by-case acceptance. It sends fully model-generated reports without human triage; reports include a self-contained reproducer, an explanation, and a candidate patch where available. Anthropic warns that findings can be invalid or have incorrect severity, and says the service is intended for projects able to handle the extra triage load. Its early review results are a limited sample, not an established accuracy rate across every project.

Official Anthropic OSS Scanner launch and methodology Anthropic

Open edition

Security · October 13, 2026

AI can speed up device configuration, but the release path must make bad changes quick to find and reverse

Fleet surveyed more than 250 enterprise IT practitioners with write or admin access to device-management tools who manage Apple devices. In its September 2026 report, 86.3% said AI-written output can reach production devices after some review; 60.7% had used AI to author MDM profiles or configurations in the prior 30 days. Separately, 69% said they could not roll back a bad configuration in under an hour, including 43% who needed more than a day. The survey is vendor-sponsored and self-reported, and it does not establish that AI use caused the rollback gap.

Primary Fleet survey of Apple-device administrators, September 2026 Fleet

Open edition

Agents · October 10, 2026

Local inference changes where a model runs, not what its agent tools can reach

Microsoft says GitHub Copilot will add automatic selection between on-device and cloud models by the end of October, alongside explicit local-model selection. Its Microsoft Execution Containers apply operating-system controls to shell commands and, by default, local MCP servers when sandboxing is enabled. Built-in file tools are checked by the Copilot harness rather than isolated as child processes, and remote MCP servers remain outside the local process sandbox. The announcement describes different enforcement paths, not a universally offline or isolated Copilot session.

Official Microsoft GitHub Copilot and Windows technical announcement Microsoft

Open edition

Security · October 10, 2026

Agent policies should consider action history and make coverage gaps explicit

AWS has released Strands Box in developer preview, starting on macOS. It combines operating-system containment with Dogwood policies evaluated at network, shell, Python, and MCP enforcement points. Those checks can use a shared action history, such as blocking outbound requests after a sensitive file read. AWS says file paths granted directly to the agent are bounded by containment but do not appear in that policy history, so policy coverage depends on how the agent reaches a resource.

Official AWS Strands Box developer-preview announcement AWS

Open edition

Security · October 9, 2026

Advanced AI-assisted cyber work is more governable when access is tied to a verified defender, an authorised scope, and monitoring proportionate to the capability

Anthropic has combined earlier cyber-access efforts into an expanded Cyber Verification Program with Defense Access, Red Team Access, and Specialized Access. It says qualifying security organisations can apply for access appropriate to defensive or authorised adversarial work; each tier has verification requirements and controls, and Anthropic retains blocks for high-risk activity such as ransomware deployment, physical-system damage, or testing high-risk safety systems.

Official Anthropic Cyber Verification Program announcement Anthropic

Open edition

Agents · October 6, 2026

A multi-cloud agent estate becomes easier to govern when every agent has a distinct identity, a task-scoped access path, and an audit trail separate from its builder or requester

IBM says its watsonx Orchestrate AI Gateway can now discover and import agents built on Microsoft Foundry and Google Gemini Enterprise Agent Platform, alongside Amazon Agentcore agents, into one control plane. Its preview Agent Identity capability assigns an agent a distinct, verifiable identity through an existing identity provider; IBM says this can support task-scoped short-lived tokens and audit records that connect the requesting user, the acting agent, and the called tool. The private preview supports IBM Verify and Microsoft Entra.

Official IBM watsonx Orchestrate announcement IBM

Open edition