Anthropic AI Updates: October 1, 2026

1. Claude for Government Reached General Availability Under FedRAMP High

Anthropic. Claude for Government, in public beta since July 2025, is now generally available to federal and state agencies in a FedRAMP High authorized environment. It includes the desktop app, Claude Code, and skills, plugins, and projects, plus admin controls such as audit logging with two-person approval for sensitive operations, SCIM-based single sign-on, MDM deployment, and conversation history kept on agency-managed devices. Pricing has no seat fees: usage is billed in fixed increments with hard spending caps, department-level budgets, and burndown alerts, while the Claude Code CLI and Claude for Microsoft 365 remain in early access. Source

2. Claude Sonnet 4.5 Will Be Retired From the Claude API on November 30

Anthropic. Anthropic announced the deprecation of Claude Sonnet 4.5 (claude-sonnet-4-5-20250929) on September 30, with retirement on the Claude API scheduled for November 30, 2026. Anthropic recommends migrating to Claude Sonnet 5.5 and published a dedicated migration path from Sonnet 4.5, which matters because Sonnet 5.5 carries breaking changes around thinking configuration, forced tool use, and older computer use tool versions. Source

3. Anthropic Built a Robot Exposure Index Covering 19,000 Job Tasks

Anthropic. Anthropic’s Economics research team used Claude to rate about 19,000 tasks across roughly 900 O*NET occupations on a four-tier scale, from tasks robots cannot do (E0) to tasks they can do in unstructured settings such as city streets (E3), and validated the index against wage and employment data going back to 1977. It found robots can technically perform 74% of physical tasks (34% of working hours) but only 2% in unstructured environments, and are cost-competitive for just 0.3% of tasks; at a historical 3% annual price decline, reaching 10% would take about 40 years. Nine of the ten most exposed occupations involve vehicle operation, and the dataset with Claude’s reasoning and cited sources for each rating was released alongside the paper. Source

4. Anthropic’s Sales Team Runs Inbound Through a Managed Agents Buying Agent

Anthropic. Anthropic described an inbound “buying agent” built by one engineer in a few weeks on Claude Managed Agents (beta), which handles hosting, sessions, and tool orchestration while the agent uses pricing and plan recommendation tools plus a knowledge base on security and compliance. The team relies on Managed Agents versioning for weekly prompt changes, staging environments, and Console editing by non-technical sales leaders. Anthropic reports thousands of conversations a day, leads converting to opportunities more than twice as often as with the previous form, deals closing about five days faster, and escalations to human reps dropping by about half. Source

5. Claude Code 2.1.286 Adds Bookmarks and Option Previews in VS Code

Claude Code 2.1.286 added a Bookmarks side panel in the VS Code extension for saving Claude’s responses, a Questions row that records the questions Claude asked and the answers picked, and previews of the highlighted option’s mockup or snippet on question cards. In the terminal, stacked permission prompts now show a count such as “2 of 5”, and fullscreen lists support clicking the “N more” rows. Admins of Claude Tag can now set spend limits on any Slack channel, including private ones, from its channel ID or link. Source