Anthropic only. Week of 2026-06-22 to 2026-06-26. 11 log entries, 7 sections. Published by id8Labs.
Tuesday, 10:02 AM ET. Every Claude model threw errors at the same instant. By 10:13 AM, outage trackers logged 7,119 simultaneous US reports. Eleven minutes of alerts. Two hours and forty-two minutes of downtime. Claude for Government never went down. No postmortem had landed by end of week. (TechRadar)
Two days later, Anthropic sent a formal letter to US senators and White House officials accusing Alibaba's Qwen AI lab of running a coordinated campaign using thousands of fraudulent accounts to extract Claude's capabilities in software engineering and agentic reasoning. (Bloomberg, CNBC) Alibaba ADRs fell to a 16-month low in Hong Kong. Anthropic has not published the letter.
The rest of the week: five Claude Code versions shipped Monday through Friday. A Slack integration put Claude inside workspaces as a named, persistent team member. The Python and TypeScript SDKs shipped in parallel, adding system-level streaming events. A research report mapped Claude usage to the rhythms of the outside world. In the background, Fable 5 and Mythos 5 remained suspended under a US export control directive in force since June 12, while Fable 5's complimentary access window on paid plans closed June 22 without a formal announcement.
Anthropic sent Congress a formal accusation: Alibaba fraudulently extracted Claude's best capabilities using thousands of fake accounts. The story is now in lawmakers' offices.
On June 24, CNBC first reported that Anthropic had sent a formal letter to US senators and the White House accusing Alibaba's Qwen AI lab of running a large-scale operation to extract Claude's capabilities using thousands of fraudulent accounts. Bloomberg confirmed and expanded the reporting on June 25. The capabilities named in the letter: software engineering performance and agentic reasoning. Alibaba ADRs fell to a 16-month low in Hong Kong trading following the disclosure. Anthropic has not published the letter. Alibaba has not commented.
The choice of venue is the story. IP theft complaints go to courts. Trade disputes with geopolitical weight go to Congress. Anthropic went to Congress. That does not make the underlying accusation false, but it tells you the company sees this as a policy argument, not a legal one.
The capabilities named in the letter are not incidental. Software engineering and agentic reasoning are the commercial core of what Claude does. Claude Code, which shipped five versions in five days this week, is the product built from those capabilities. The week where Alibaba was accused of fraudulently acquiring them was also the week Anthropic shipped the most rapid public iteration of the product that runs on them.
The political context matters. Fable 5 and Mythos 5 have been off the API since June 12, suspended following a public jailbreak demonstration and a US government export control directive. Anthropic published a statement disagreeing with the restriction but complying. (Anthropic statement) The letter to Congress creates a coherent posture: a company that accepted external constraints on its own frontier models while asking government to constrain those who would extract those models' capabilities through fraud.
Expect regulatory coverage. The story is in the pipeline of multiple lawmakers' offices.
Fable 5 and Mythos 5 remain off the API. The US government export control directive, issued June 12 after a jailbreak was publicly demonstrated on X, was not lifted this week. Anthropic published a statement calling the action a response to a narrow potential jailbreak, disagreeing with it, and complying. (Anthropic statement) No new developments confirmed in the weekly window.
Fable 5 complimentary access ended June 22. The no-extra-cost window on Pro, Max, Team, and seat-based Enterprise plans, in place since June 9 (the day of the model launch), closed Monday. Access reverted to standard plan entitlements. No formal announcement of the transition was published.
A 1:1 mirror of every Anthropic release in the window, June 22 to 26, 2026. Use it as reference. 11 entries, 6 categories.
Five versions in five days. Monday through Friday, one per day. MCP CLI auth, a /rewind command, auto-mode shell routing, OpenTelemetry logging, and voice dictation fixes.
claude mcp login <name> and claude mcp logout <name>, with --no-browser for SSH sessions. Bash ! commands now trigger Claude auto-response by default. iTerm2 teammate mode added (teammateMode: "iterm2"). AWS credential refresh option in /login for awsAuthRefresh setups. The agent gets a compaction reminder when MEMORY.md approaches its size limit. Fixes: streaming after machine sleep, subagent transcript rendering, background task previews, permission prompt alignment, workflow schema loops capped at 5 attempts, YAML frontmatter handling across kebab-case and snake_case.claude update or reinstall. Bash auto-response is on by default; add "respondToBashCommands": false to settings to revert. For SSH: claude mcp login <name> --no-browser and follow the stdin prompt.
/rewind restores conversation context from before a /clear command was run. Streaming CPU usage drops approximately 37% through text update coalescing at 100ms intervals. Long-session memory growth from terminal output caching is reduced. MCP reliability improves with retry logic on transient network errors for capability discovery and OAuth flows.claude update or reinstall. /rewind is available in any session that had a prior /clear.
autoMode.classifyAllShell setting routes all Bash and PowerShell commands through the auto-mode classifier, not just arbitrary-code-execution patterns. New claude_code.assistant_response OpenTelemetry log event emits model response text as a structured log, redacted by default and opt-in via OTEL_LOG_ASSISTANT_RESPONSES=1. Live file path autocomplete in bash mode (! prefix). Idle background shell commands now auto-reap under memory pressure. Three backgrounding bugs fixed: spurious cancellation warnings on exit, phantom "general-purpose (resumed)" subagent spawned on main turn backgrounding, and pinned agents being re-prompted after auto-update.claude update. Set autoMode.classifyAllShell: true in settings to route all shell commands through the classifier. Set OTEL_LOG_ASSISTANT_RESPONSES=1 to enable response logging.
CLAUDE_CODE_DISABLE_MOUSE_CLICKS environment variable disables click interactions in fullscreen mode while preserving wheel scroll. Hook matchers with hyphenated identifiers now use exact matching instead of substring matching. Voice dictation on macOS is repaired, including auto-submit for languages without spaces. Linux voice mode improved. Remote session startup adds a provisioning checklist.claude update or reinstall from the GitHub releases page. Set CLAUDE_CODE_DISABLE_MOUSE_CLICKS=1 to disable click interactions while keeping scroll.
system.message streaming events: clients can receive streamed system-level events mid-conversation alongside content_block_* events. Also adds a new refusal category to the stop reason taxonomy and User Profile ID support in request headers. Bug fix: the memory tool now creates parent directories with correct permissions.pip install --upgrade anthropic. Handle system.message events in your stream loop alongside existing content_block_* events.
system.message streaming events, the new refusal category, and User Profile ID in request headers. Fixes the x-stainless-helper header to use a single source with append semantics and correct fallback middleware tagging.npm install @anthropic-ai/sdk@latest
Daily digest at 9 PM ET. Weekly magazine every Friday morning. Six labs, one feed. No spam, one-click unsubscribe.