Anthropic AI Updates: July 8, 2026

1. Anthropic Brings Claude Code and Claude Cowork to Government in Public Beta

Anthropic. Anthropic announced public beta availability of Claude Code and Claude Cowork within Claude for Government Desktop, delivered through a FedRAMP High authorized environment. The release adds government-specific governance controls including administrative permissions, audit logging, and billing tied to government appropriations. The aim is to let public-sector agencies modernize legacy systems and delegate administrative work under compliance oversight. Source

2. Claude Cowork Expands to Mobile and Web

Anthropic. Anthropic is extending Claude Cowork beyond desktop to mobile and web, letting users start a task on one device and continue it on another while work proceeds remotely in the background. The rollout introduces three behavioral changes: work state follows the user across devices, scheduled tasks execute in the background without a device online, and Claude pauses to request explicit approval before important decisions. The update pushes Cowork toward an always-on assistant model rather than a single-session desktop tool. Source

3. Anthropic Details How People Are Using Claude Cowork

Anthropic. Anthropic analyzed roughly 1.2 million sampled Claude Cowork sessions from May and found that about half of usage falls into “the work around the work,” administrative tasks like status reports, slide decks, and spreadsheet organization that cut across job functions. Business process and operations tasks accounted for the largest share at 33.4 percent, followed by content creation at 16.4 percent. The pattern suggests knowledge workers primarily delegate coordination and communication overhead rather than core domain expertise. Source

Anthropic. Claude Code releases v2.1.202 and v2.1.203 add several developer-facing features, including a warning before login expiry so background sessions are not interrupted, a grey pause badge in the footer to indicate manual permission mode, and additional session working directories exposed via MCP’s roots/list with change notifications. The releases also add a “Dynamic workflow size” setting in /config to control how large Claude typically makes dynamic workflows, plus workflow.run_id and workflow.name OpenTelemetry attributes on telemetry from workflow-spawned agents. Source

5. Anthropic Publishes Guidance on Model Choice Versus Effort Level in Claude Code

Anthropic. Anthropic published guidance distinguishing model selection from effort-level configuration in Claude Code, clarifying that the model choice determines which trained weights handle a request while effort level governs how thoroughly Claude works, including how many files it reads, tests it runs, and verification steps it performs. The guidance recommends sticking with default effort levels for most tasks and only adjusting model or effort when Claude either lacks the capability for a complex problem or under-invests in a routine one. Source

6. Anthropic Publishes an Oral History of Claude Code

Anthropic. Anthropic published a feature-length retrospective tracing Claude Code’s evolution from an internal command-line tool used by its own researchers into the company’s flagship coding agent. The piece is told through interviews with the researchers, engineers, and early internal users who built and iterated on the product. It covers the design decisions and early adoption patterns that shaped its current form. Source