Mythos cleared for US release

Highlight

Rewind Now Works Across /clear

You can now use /rewind to step back into a conversation from before you ran /clear, recovering context you previously had to give up to free the window. If you regularly clear mid-session to refocus Claude, this removes the cost of doing so prematurely.

Worth Knowing

Bash Mode Now Prompts Claude on Output

Running a ! bash command now automatically asks Claude to respond to the output instead of just adding it to context. Set "respondToBashCommands": false in settings.json if you preferred the old behavior of using ! purely to feed context.

File Path Autocomplete Comes to Bash Mode

Bash mode now offers live file path completion as you type, matching the behavior you already get in the main prompt. Useful if you frequently chain !ls, !cat, or !grep against paths you don't want to retype.

Authenticate MCP Servers From the CLI

You can now run claude mcp login <name> and claude mcp logout <name> without entering the interactive /mcp menu, and --no-browser redirects the flow through stdin for completing OAuth over SSH. Helpful if you manage MCP servers on remote machines or in scripts.

Background Subagents Now Ask Before Acting

Background subagents no longer auto-deny tool requests that need permission. The prompt now surfaces in your main session showing which agent is asking, and pressing Esc denies only that single tool call rather than killing the agent.

Sandbox Can Block Credential Reads; Model Restrictions Surface in the Picker

A new sandbox.credentials setting blocks sandboxed commands from reading credential files and secret environment variables. Separately, org-configured model restrictions now appear directly in the model picker, --model, /model, and ANTHROPIC_MODEL, with a clear message when you select a restricted model.

Under the Hood

37% Less CPU During Streaming, Lower Long-Session Memory

Streaming responses now coalesce text updates to 100ms, cutting CPU by roughly 37%, and the terminal output cache no longer drives long-session memory growth.

Hook Matchers, Background Agents, and VS Code Resume Stabilized

Hyphenated hook matchers like code-reviewer and mcp__brave-search now exact-match instead of substring-matching, comma-separated matchers like "Bash,PowerShell" actually fire, stopped background agents stay stopped, and the VS Code extension no longer hangs when resuming a large session.

From Anthropic

Anthropic Engineering · 10 min read

Building effective human-agent teams

Anthropic walks through patterns for working alongside multiple agents at once, with examples drawn from real workflows. Relevant if you've been experimenting with the agent teams and background agent features that have landed over the last several releases.

Anthropic Engineering · 10 min read

Building effective human-agent teams

Anthropic walks through patterns for working alongside multiple agents at once, with examples drawn from real workflows. Relevant if you've been experimenting with the agent teams and background agent features that have landed over the last several releases.

Anthropic News · 3 min read

Introducing Claude Tag

Anthropic introduces Claude Tag, a Slack-based way for teams to bring Claude into shared channels with a new agent identity access model. Worth a look if your team coordinates engineering work in Slack and you've been wondering how Claude fits into team-wide threads.

Anthropic News · 3 min read

Introducing Claude Tag

Anthropic introduces Claude Tag, a Slack-based way for teams to bring Claude into shared channels with a new agent identity access model. Worth a look if your team coordinates engineering work in Slack and you've been wondering how Claude fits into team-wide threads.

In the News

Wired · 5 min read

Trump Administration Allows Anthropic to Release Mythos to Select US Organizations

Following last week's coverage of negotiations, the Trump administration has agreed to let Anthropic release Mythos to a vetted list of US companies and agencies. If your team's model access has been affected by the earlier restrictions, this is the resolution to track.

Wired · 5 min read

Trump Administration Allows Anthropic to Release Mythos to Select US Organizations

Following last week's coverage of negotiations, the Trump administration has agreed to let Anthropic release Mythos to a vetted list of US companies and agencies. If your team's model access has been affected by the earlier restrictions, this is the resolution to track.

Wall Street Journal · 5 min read

Anthropic Claims Alibaba Ran 'Brazen' Campaign to Access Its Claude AI Model

Anthropic accuses Alibaba of running a sustained campaign to illicitly extract Claude's capabilities through distillation. The dispute is reshaping how Anthropic talks about model access controls and downstream policy for API customers.

Wall Street Journal · 5 min read

Anthropic Claims Alibaba Ran 'Brazen' Campaign to Access Its Claude AI Model

Anthropic accuses Alibaba of running a sustained campaign to illicitly extract Claude's capabilities through distillation. The dispute is reshaping how Anthropic talks about model access controls and downstream policy for API customers.

Wall Street Journal · 6 min read

China Has Matched Anthropic in Cybersecurity, Resetting AI Race

The Journal reports that Chinese labs have closed the gap with Anthropic on cybersecurity-relevant capabilities, including those that triggered the Mythos restrictions. Useful framing for understanding why the export and access debates are escalating.

Wall Street Journal · 6 min read

China Has Matched Anthropic in Cybersecurity, Resetting AI Race

The Journal reports that Chinese labs have closed the gap with Anthropic on cybersecurity-relevant capabilities, including those that triggered the Mythos restrictions. Useful framing for understanding why the export and access debates are escalating.

The Information · 5 min read

How AI Customers Are Lowering Their Anthropic and OpenAI Bills

The Information looks at how heavy API users are shifting toward efficiency-focused workflows, including smaller models, prompt caching, and routing. Relevant if you're trying to keep Claude Code costs predictable as usage grows.

The Information · 5 min read

How AI Customers Are Lowering Their Anthropic and OpenAI Bills

The Information looks at how heavy API users are shifting toward efficiency-focused workflows, including smaller models, prompt caching, and routing. Relevant if you're trying to keep Claude Code costs predictable as usage grows.