From frontier-model launches and managed agent runtimes to smart-home control and independent oversight, the period since our last edition brought major developments across the AI-agent stack.
Here’s everything worth catching up on in the latest edition of AI Agents Pulse:
1️⃣ OpenAI released GPT-6 Astra, its most capable model to be broadly deployed and the first OpenAI model to reach the Critical level of cybersecurity capability under its Preparedness Framework.
https://deploymentsafety.openai.com/gpt-6-astra
2️⃣ Google introduced Gemini 3.8 Flash for long-horizon coding and autonomous agents at $0.75 per million input tokens and $3.75 per million output tokens, alongside Gemini 3.8 Flash Cyber for trusted defenders through the Fairwind Program.
https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/
3️⃣ OpenAI released the Agents API in public beta for all developers. Powered by the open-source Codex harness, it supports long-running sessions, MCP and other tools, and multi-agent orchestration with subagents.
https://openai.com/index/introducing-the-agents-api/
4️⃣ Meta introduced Muse, a personal agent running in a dedicated Muse Secure VM. It can browse, fill forms, keep working after the app closes, and seek approval before sending email or making purchases.
https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/
5️⃣ Anthropic announced a redesigned Claude Code Projects beta whose coordinator scopes work, delegates to parallel cloud threads, reviews outputs, and shares project memory. Initial access is limited to select Pro and Max subscribers using cloud sessions.
https://claude.com/blog/projects-redesigned
6️⃣ Google made Home MCP available in early access, letting compatible agents inspect home structures and device state, control devices, and query history. Its safeguards prohibit sensitive actions such as unlocking doors.
https://developers.home.google.com/mcp/home
7️⃣ Salesforce introduced AIforce, a live interface layer that lets agents reason and act across Salesforce data, workflows, business logic, permissions, security, and governance from interfaces including Claude and Slack.
https://www.salesforce.com/uk/news/stories/aiforce-announcement/
8️⃣ Anthropic announced a non-exclusive embedded-evaluation partnership with Accenture, led by Faculty. Each company expects to invest at least $1 billion over five years, while operating details and evaluation standards are still being worked out.
https://www.anthropic.com/news/accenture-embedded-evaluation
9️⃣ California Governor Gavin Newsom issued an executive order accelerating independent AI oversight and directing experts to deliver recommendations within two months, including how to advance an independently verified kill switch for frontier models. The order seeks recommendations rather than imposing a completed kill-switch mandate today.
https://www.gov.ca.gov/2026/09/18/governor-newsom-issues-executive-order-to-accelerate-independent-oversight-and-advance-the-creation-of-an-ai-kill-switch/