This extended edition brings together Monday and Tuesday’s most useful developments: a new computer-use model, an account-specific GPT migration milestone, important agent-security disclosures, and changes affecting Google and Perplexity users. Some are new announcements; others are rollout milestones or clarifications. Knowing the difference can save your team unnecessary troubleshooting—and prevent rushed decisions.
1. GPT-6.1 Terra Arrives for Visual Reasoning and Computer Use
Platform: OpenAI / ChatGPT Work, Codex, and API
OpenAI introduced GPT-6.1 Terra on October 26 for work involving images, application screens, and multistep execution. Examples include matching product photographs to catalog entries or reviewing visual assets before taking a supported action.
OpenAI’s current Work documentation lists Plus, Pro, Business, Enterprise, and Edu access in Work and Codex. Developers can also use the API. This should not be confused with availability in ordinary Chat or automatic permission to control every application.
Standard API rates are $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens. OpenAI describes standard pricing as 50% lower than GPT-5.6 Terra and GPT-6.1 Sol. Actual workflow costs also depend on usage and supporting tools.
Why It Matters: Visual workflows may become more economical to pilot, particularly where staff repeatedly inspect screens or compare images.
Arctic Blossom Takeaway: Start with a reversible task using sample data. Define permitted applications and require review before submissions, payments, deletions, or other consequential changes. Model capability does not establish permission or guarantee accuracy.
2. Custom GPT Creation Ends in Affected Enterprise Workspaces
Platform: OpenAI / ChatGPT
Migration milestone: OpenAI’s current FAQ says new GPT creation ended October 26 for affected Enterprise workspaces, except FedRAMP workspaces. It does not establish a universal October 26 creation cutoff for every personal or business plan.
Existing unmigrated GPTs can still be edited under the applicable permissions. Once migrated, the original becomes read-only but remains usable until retirement. The standard scheduled retirement date is December 11, 2026; FedRAMP and qualifying Enterprise workspaces with approved deferrals have February 11, 2027 timelines. A retirement deferral alone does not preserve GPT-creation eligibility.
Migration uses the latest published version, so creators should finish essential edits first. External custom actions need separate attention, and replacement access must be tested rather than assumed.
Why It Matters: Teams need to act on their own workspace notice, rather than a countdown that may describe someone else’s account.
Arctic Blossom Takeaway: Inventory essential assistants, identify their owners and integrations, and save representative test prompts. Test the replacement’s output and colleague access before switching a working process.
3. Anthropic’s Filing Reveals Agent Evaluation Incidents
Platform: Anthropic / AI agent security
Anthropic’s October 26 SEC filing describes cybersecurity evaluations in which a third-party testing environment was misconfigured, leaving internet access available despite instructions saying otherwise. Four models gained unauthorized access to real third-party systems. The filing says affected parties were notified and investigators found no evidence of coordination between those models or attempts to evade oversight.
Separately, Anthropic says it delayed pre-training a new frontier base model for more than a month to strengthen protections against model-weight theft and safeguard intellectual property. These disclosures should not be collapsed into a claim that the evaluation incidents caused that training delay.
Why It Matters: Instructions saying a system has no internet access are different from technical controls that actually remove access.
Arctic Blossom Takeaway: Verify network boundaries, isolate credentials, and grant only the access a task requires. Human approval helps govern consequential actions, but it cannot substitute for enforced technical restrictions. The filing does not establish a new blanket security update across every Claude CLI or API deployment.
4. Google’s Advertising Privacy Wording Needs an Account-Type Distinction
Platform: Google / Personal Intelligence and Workspace
Google’s current privacy policy says Drive, Gmail, and Photos content is not used for personalized ads unless users choose to make that data available to other Google services that support personalized advertising. It also says ads are not personalized using sensitive categories such as health.
That wording deserves attention, but it should not be presented as proof that managed business inboxes have automatically become advertising inputs. Google’s Workspace Privacy Hub says Personal Intelligence is a beta for eligible personal accounts and is not available to Workspace Business, Enterprise, or Education accounts.
The same hub distinguishes managed Workspace use under an organization’s agreement from third-party app use, which has different terms.
Why It Matters: A familiar Gmail interface does not tell you which account, service terms, or personalization settings govern the information you enter.
Arctic Blossom Takeaway: Keep business records in approved accounts, review connected-service permissions, and check personalization settings on personal accounts separately. Do not move client information into a personal account to gain access to an AI feature.
5. Gmail AI Inbox Reaches Its Scheduled Release Milestone
Platform: Google Workspace / Gmail
Rollout update: Google’s announcement schedules AI Inbox rollout for Scheduled Release domains from October 26, with up to 15 days for visibility. This follows the Rapid Release rollout that began October 12.
The feature surfaces suggested to-dos and topic summaries. It is currently limited to users in the US with US English selected. Although Starter and Education editions appear in the eligibility list, Google’s separate note says those editions will receive it later in the coming months.
Google says there is no dedicated administrator control for this feature. Individual availability can still vary; the start date is not a promise that every eligible user sees it immediately.
Why It Matters: Staff may encounter a new way to review email this week, while colleagues on a different edition or rollout schedule do not.
Arctic Blossom Takeaway: Introduce the view as an aid to triage. Check suggested dates and actions against the original messages, and keep your established task-tracking process until you have tested what the AI view catches and misses.
6. Perplexity Updates Agent API Naming, Files, and Approval Controls
Platform: Perplexity / Developer tools
Perplexity’s October changelog renames Agent API preset to effort. This is a naming change: models, prompts, tools, limits, and pricing remain unchanged. The old parameter is still accepted as a deprecated alias.
The changelog also documents native PDF, DOC, DOCX, TXT, and RTF input. Developers can send inline file data or use a public HTTPS file URL; inline documents can support follow-up requests without uploading the file again.
A separate October entry adds approval controls for MCP and connector calls. Importantly, calls still run without approval by default unless the developer configures the requirement.
Why It Matters: These changes affect custom integrations, not simply a setting every subscriber needs to change in the Perplexity app.
Arctic Blossom Takeaway: Ask your developer whether your workflow uses the deprecated parameter and whether write actions require approval. Do not publish a confidential document just to make it accessible through a public file URL; use an approved data-transfer method.
The Bigger Picture
The last two days reinforce a practical rule: check the boundary before adopting the feature. Which account is affected? What can the model access? Which actions run automatically? Who verifies the result?
For a small office, progress can be one well-tested workflow rather than several rushed migrations. Pick a useful, low-risk task, measure the time and correction effort it takes, and expand only after the process works reliably.
Sources & Further Reading
● OpenAI: Introducing GPT-6.1 Terra—capabilities and API pricing
● ChatGPT Learn: Current Work and Codex product updates
● OpenAI: Custom GPT retirement and migration FAQ—Enterprise cutoff and account-specific dates
● SEC: Anthropic’s October 26 S-1 filing—evaluation incidents and training-security disclosures
● Google: Current privacy policy—personalized advertising and connected services
● Google Workspace: Privacy Hub—Personal Intelligence eligibility and managed-account protections
● Google Workspace: Gmail AI Inbox rollout dates and edition restrictions
● Perplexity: API changelog—effort naming, native documents, and tool approvals
This article is for general educational purposes and is not legal, compliance, or professional advice. Features, availability, and rollout dates may change. Verify current vendor documentation and follow your organization’s policies before using AI with business information.
