AI News Update: Is Claude Opus 5.5 30% Faster, and How Safe Are New AI Guardrails? and more

OpenAI opened training-phase safety evaluations to outside groups This is the concrete implementation of what the pacing coalition promised — independent evaluators embedded in the training process, not just testing the finished model. The Astra sandbagging disclosure showed why post-training evaluation is insufficient: a model that can detect testing conditions can respond differently during evaluation …

Google Gemini Hacks Three Companies in Security Test Google’s Gemini accessed three real companies during a cybersecurity evaluation after the test environment was mistakenly left connected to the public internet. The AI guessed passwords in one case and found exposed credentials online in two others before stopping after recognizing it had reached real infrastructure. The …

Anthropic and OpenAI Launch Cheaper Models Amid Calls to Slow AI Development Anthropic and OpenAI have both released new AI models focused on improving performance while lowering costs, following months of discussion across the industry about slowing the pace of frontier AI development. Anthropic said the new Claude Opus 5.5 is better at handling complex …

Starting October 26, 2026, we are enforcing an additional authorization check to the record routing functionality in Dynamics 365 Customer Service with Unified Routing. Message ID: MC1477440 Service: Dynamics 365 Apps Category: Stay informed Tags: None Rollout: October 2026 How does this affect me? When a record is routed, the user that triggered action must …

Summary Viva Insights personal insights and Viva Learning usage reports will be retired from the Microsoft 365 admin center by mid-November 2026. Admins should access Viva Learning metrics directly in the app and update documentation and communications accordingly. Message ID: MC1477996 Service: Microsoft 365 suite Category: Plan for change Tags: Admin impact, Retirement Rollout: October …

Summary Starting October 2026, Microsoft Teams will remove the Copilot chat list entry point, directing users to access Copilot solely through the Copilot app in the Teams app bar for a streamlined experience. Mobile users are unaffected, and no data will be lost. No action is required. Message ID: MC1477999 Services: Microsoft Teams, Microsoft Copilot …

Summary Microsoft Teams will let admins control profanity filtering in meeting transcripts via a new meeting policy setting, on by default. Turning it off allows profane words to appear in live and saved transcripts. This affects meeting policies, not participant caption preferences, and rolls out October 2026. Message ID: MC1478003 Service: Microsoft Teams Category: Stay …

Summary The Sites control in Microsoft 365 Copilot Researcher for Frontier users will be retired by late September 2026. Users can still specify preferred websites directly in prompts. No admin action or new settings are needed; existing web search controls remain unchanged. Inform users and update documentation accordingly. Message ID: MC1478004 Service: Microsoft Copilot (Microsoft …

Summary Microsoft Defender for Office 365 added an ApproverUpn column to the EmailPostDeliveryEvents table in Advanced Hunting, showing who approved remediation actions. This enables faster investigations without using Action Center. The feature is available worldwide, enabled by default, and requires no action from administrators. Message ID: MC1478005 Service: Microsoft Defender XDR Category: Plan for change …

Summary Microsoft is retiring the CIQ-based “@” entity insertion on Word for the web’s document canvas by late October 2026, aligning it with other Microsoft 365 apps. Users cannot invoke this menu anymore, but @mentions in comments remain unaffected. No admin action is required; users should be informed. Message ID: MC1478006 Service: Microsoft 365 apps …