China Alleges Claude Code Vulnerabilities
Claude Opus 4.8 is now the default in Claude Code across Bedrock, Vertex AI, and Claude Platform, launching with GA Auto mode and an upgraded /doctor that diagnoses and fixes setup issues. In other news, China alleged Claude Code vulnerabilities, Ben Bernanke joined Anthropic's Long-Term Benefit Trust, and TeraWulf signed a $19 billion infrastructure lease with Anthropic.
China Says It Has Found Security Vulnerabilities in Anthropic's Claude Code — Wall Street Journal
Highlight
Claude Opus 4.8 Becomes Default on Cloud Providers
If you run Claude Code through Bedrock, Vertex AI, or Claude Platform on AWS, your sessions now default to Claude Opus 4.8. You can still switch via /model, but expect a different baseline for cost, latency, and output quality on those providers going forward.
Worth Knowing
Auto Mode Goes GA on Bedrock, Vertex, and Foundry
Auto mode no longer requires the CLAUDE_CODE_ENABLE_AUTO_MODE opt-in on Bedrock, Vertex AI, and Foundry. If your organization needs it off, set disableAutoMode in settings. Note that autoMode is no longer read from .claude/settings.local.json -- move it to ~/.claude/settings.json.
/doctor Becomes a Full Setup Checkup
/doctor (aliased as /checkup) now actively diagnoses and offers to fix installation and configuration issues rather than just reporting them. It also proposes trimming checked-in CLAUDE.md files by cutting content Claude could derive from the codebase, and reports when a custom launcher script or symlink at ~/.local/bin/claude is externally managed.
Background Agents Get Smarter Lifecycle Handling
Background agents now upgrade in the background right after a Claude Code update, so attaching to a stale session no longer pays a slow upgrade cost. You'll also see a warning before your login expires so background sessions aren't interrupted, and the agent view now links PRs that a session edited, merged, commented on, or pushed to.
/commit-push-pr Now Respects Your Configured Push Remote
If your repo uses remote.pushDefault or a single non-origin remote, /commit-push-pr will auto-allow git push to it without prompting. This removes friction for fork-based workflows and repos where origin isn't the push target.
/review Reverts to a Single-Pass Review
/review <pr> is once again a fast single-pass review. If you want the multi-agent review at a chosen effort level, use /code-review <level> <pr#> -- and note that /code-review findings quality has improved on claude-opus-4-8 across all effort levels.
Under the Hood
Terminal Responsiveness and Streaming Fixes
This week fixes the terminal freezing on very long streamed responses, transcript jumping when a response finishes, content jumping while scrolling long transcripts, and claude --resume/--continue not responding to keyboard input on startup. Interactive sessions also stop re-analyzing the entire transcript after every turn, cutting per-turn CPU and memory.
Plugin Shell-Injection Fix and Updater Memory Cut
${user_config.*} in shell-form plugin hook, monitor, and MCP headersHelper commands is now rejected to prevent shell injection -- use exec form (args array) or $CLAUDE_PLUGIN_OPTION_<KEY> instead. Also fixed: remote managed settings from non-interactive runs being silently recorded as consented, and the auto-updater now streams downloads to disk, cutting roughly 400 MB off peak memory usage.
From Anthropic
In the News
A quieter week on new features, but the Opus 4.8 default on cloud providers and the auto mode GA are meaningful shifts if you run Claude Code through Bedrock, Vertex, or Foundry.