ainewsfab
Anthropic Says Claude Leads 26% of Its AI R&D—But the Metric Is Still Unaudited
Anthropic says Claude leads 26% of measured AI R&D. Here is what its automation, agent-monitoring, and safety-compute metrics do—and do not—show.
Kiro’s Supervised Diff Was Not a Firewall. AWS Says Upgrade and Rotate Credentials.
AWS says older Kiro IDE versions could expose workspace data through an agent-written setting before a pending diff was resolved. Here is who should upgrade and rotate credentials.
Apple’s Siri AI Arrives With a Bigger Promise—and Clearer Limits
Apple has begun rolling out Siri AI in beta as part of its 2027 software releases, turning the company’s familiar voice assistant into a more conversational system that can use personal context, understand what is on screen, search the web, and take broader actions across apps. The shift matters less because it adds another chatbot…
AWS Wants Agents to Propose. Step Functions Still Gets to Decide.
AWS’s reference design for airline rebooking separates agent proposals from deterministic validation and execution. The useful idea is broader than travel: an agent should not be the final authority for an irreversible business action.
DeepSeek V4.1-Flash Makes AI Cost a Cache Problem
DeepSeek’s new open-weight model promises to shrink the memory bill for long-running agents. The engineering claim is meaningful, but buyers still have to price the whole workflow—not just the token line item.
Gemini Lands on Windows, but the Real Product Is Access—Not Local AI
Google’s new Gemini Windows app makes cloud AI a keyboard shortcut. Here is what it requires, what Workspace admins control, and why a desktop client is not local AI.
OpenAI’s GPT-Live-1 Makes Voice Agents Less Awkward. The Hard Part Is Still the Call.
OpenAI has made GPT-Live-1 generally available in its API, and its most consequential feature is not a new voice. It is the refusal to stop listening while the rest of the system thinks. Traditional voice agents tend to move a call through a relay race: speech-to-text, a language model, then text-to-speech. A caller pauses, interrupts,…
Google Cloud Gives Coding Agents a Portable Plugin—With IAM Guardrails Attached
Google has packaged Cloud knowledge and actions for coding agents. The practical benefit is less hand-wired setup; the operational risk is treating a plugin install as proof of least privilege, current documentation or a traceable infrastructure change.
Salesforce Says the AI Model Is Not the Moat. Its Rollout Schedule Says Otherwise.
Salesforce’s new Enterprise AI Harness promises one control layer for agents, models and business actions. The unified experience is still a FY28 plan, while pricing remains open.
OpenAI’s OneGov Deal Makes AI Cheaper to Start—Not Easier to Budget
The new 27-month GSA agreement drops the user license fee to zero and halves usage costs. The missing number is the one procurement teams need most: the actual rate card.