Archive — 2026-06Week 4

Fortune: Anthropic's Alibaba distillation fight raises a trillion-dollar question for its IPO — can a frontier AI moat hold against China? — Fortune published an in-depth analysis on 2026-06-28, following up on Anthropic's 24 June letter to the US Senate accusing Alibaba of using roughly 25,000 fake accounts and over 28.8 million interactions to distill Claude, and examining what the case means for Anthropic's IPO, which could value the company at $1 trillion. IPO expert Jay Ritter laid out two opposing views: on one hand it cements Anthropic's strategic role in US-China tech rivalry and strengthens its IPO story; on the other, if its frontier AI moat cannot hold, investors may question future profitability ("how much they'll be able to sustain that is a big question mark"). The article highlights a regulatory gap — current export controls target physical hardware and tangible software, not API-based distillation; former assistant commerce secretary Kevin Wolf said plainly that "Querying it through an API is not exporting," so such activity escapes existing controls. On the legislative response, the Remote Access Security Act would broaden export controls to cover foreign cloud access, with Rep. Michael Lawler (R-New York) saying "Dangerously, that's exactly what's happening right now with Alibaba." The piece also cites PitchBook senior analyst Harrison Rolfes.

Reuters: Austria lobbies EU to host Anthropic in Europe to counter US AI export controls — Reuters reported on 2026-06-28 that Austria's State Secretary for Digitalisation Alexander Pröll sent a letter to EU technology commissioner Henna Virkkunen, urging member states to “jointly explore the strategic establishment and participation of Anthropic within the European Union” so that Europe is not cut off from major AI innovations and to strengthen technological autonomy. It is the first EU-level institutional response after the 12 June US export-control directive ordered Anthropic to suspend access to Claude Fable 5 and Mythos 5 for all non-US nationals (leaving European businesses, researchers and government agencies without access overnight). The European Commission has said it is in dialogue with Anthropic about possible operational adjustments, framing it within its pursuit of technological autonomy.

Claude Opus 4.8 elevated error rate incident (resolved) — status.claude.com logged an incident affecting Claude Opus 4.8 (titled 'elevated errors on Opus-48'): at 2026-06-27 18:00 UTC it was reported as mitigated and under monitoring (the status page shows only a single update with no separate onset time). This continues the pattern of recurring frontier-model error-rate incidents since Opus 4.8 launched.

Axios scoop: US government close to restoring public access to Claude Fable 5, restrictions could lift as soon as next week — Axios reported exclusively on 2026-06-27 (syndicated via Reuters) that, following the Commerce Department's 6/26 approval to redeploy Mythos 5 to over 100 trusted US organizations, the Trump administration is moving toward restoring general public access to Fable 5, with insiders expecting the restrictions to be lifted as soon as next week and negotiations continuing over the weekend. Fable 5 has been offline for 15 days since the 6/12 export control directive. Officials said the government is 'moving toward allowing Anthropic to release Fable soon, although a timeline is unclear,' and final clearance still requires sign-off from the Pentagon and the National Security Agency (NSA), leaving the outcome uncertain. Anthropic International Managing Director Chris Ciauri earlier said he is 'very confident that in the coming days the models will become available again.' Anthropic and the White House did not immediately respond to requests for comment.

Claude Code v2.1.195 released — adds `CLAUDE_CODE_DISABLE_MOUSE_CLICKS` to disable mouse click/drag/hover in fullscreen mode while keeping wheel scroll. Bug fixes: hook matchers with hyphenated identifiers (e.g. `code-reviewer`, `mcp__brave-search`) that accidentally substring-matched now exact-match (use `mcp__brave-search__.*` to match all tools from a hyphenated MCP server); macOS voice dictation capturing silence in long-running sessions after the default input device changes; voice dictation auto-submit never firing for languages written without spaces (Japanese, Chinese, Thai); external plugins enabled only by project `.claude/settings.json` not requiring explicit install consent on every loader path; `/plugin` Enable/Disable failing when a plugin's `plugin.json` `name` differs from its marketplace entry name; background jobs disappearing from `claude agents` or losing data when written by a newer Claude Code version; reopening a crashed background task showing a blank screen for up to 5 seconds instead of its restart; background agent daemons running unreachable when the control socket fails to start, blocking restarts. Improvements: Linux voice mode now distinguishes 'no microphone' from 'SoX not installed' when SoX is present but no capture device exists; `claude agents` completed list fills available vertical space (the header compacts on short terminals so live sessions stay visible); Remote session startup shows a provisioning checklist while the container starts.

TechCrunch/NBC/CNN: US Commerce Department partially lifts export controls, allowing Anthropic to restore Claude Mythos 5 to over 100 'trusted' US organizations; Fable 5 remains restricted — TechCrunch (reporter Julie Bort) and multiple outlets reported on 2026-06-26 that US Commerce Secretary Lutnick wrote to Anthropic (addressed to Tom Brown), stating he had 'determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model,' permitting Mythos 5 to be redeployed to over 100 US government agencies and companies that operate and defend critical infrastructure, and restoring access for these approved organizations (including Anthropic's own non-American employees). This is a partial reversal after roughly two weeks of negotiations following the 6/12 export control directive that required Anthropic to disable Fable 5 and Mythos 5 for all non-US nationals; Lutnick's letter is silent on Fable 5, which has not been cleared to return to the general market. Mythos 5 is Anthropic's strongest cybersecurity model, previously offered to trusted organizations such as Cisco and JPMorgan Chase through Project Glasswing. Anthropic posted on X that it is 'restoring access for these organizations quickly' and will keep negotiating with the government to expand Mythos 5 access and make Fable 5 available for general use again.

Anthropic Economic Index report 'Cadences' — revealing daily and weekly rhythms of economic life through Claude usage — On 2026-06-26 Anthropic published its latest Anthropic Economic Index report, 'Cadences,' analyzing how Claude usage reflects daily and weekly rhythms of economic activity. Key findings: personal-use conversations rise from about 35% on weekdays to about 50% on weekends, while work-related tasks decline outside business hours; usage follows predictable daily cycles (news queries peak at 7 a.m., recipe requests are 2.3x more frequent at 6 p.m., sleep advice peaks around 5 a.m.); higher-wage occupational tasks make up a larger share of work at night and on weekends, suggesting professionals keep more flexible hours; 93% of conversations produce an identifiable artifact, most commonly explanations (17%), documents/reports (15%), and guidance (11%). The accompanying survey also found over a third of respondents expect AI to handle most of their work tasks within 12 months, and those delegating the most to Claude are the most optimistic about career impacts.

CNBC: enterprises shift from 'tokenmaxxing' to cost efficiency; AI startup Lindy moves 100% of traffic off Claude to DeepSeek — CNBC reported on 2026-06-26 that as OpenAI and Anthropic both filed confidentially for IPOs in early June at valuations approaching $1 trillion, business attitudes toward AI spending are shifting from 'spend-at-all-costs' to demanding clear ROI. AI startup Lindy (CEO Flo Crivello, ~25 people) moved 100% of its traffic this month off Anthropic's Claude models to Chinese company DeepSeek's cheaper open-weight models, saying the cost curve 'crashed to the ground' and will save millions within months — 'a matter of survival for the business' — though he'd switch back if Anthropic cuts prices. Anthropic reported a $47 billion annualized run rate in May (up from ~$10 billion in revenue all of last year). D.A. Davidson analyst Gil Luria said Anthropic/OpenAI's 'current growth rates are the fastest they will ever be,' making it strategic to go public before large enterprise customers start reining in out-of-control token spend. The piece also cites Uber burning through its entire annual AI budget in four months and now imposing spending tiers starting at $1,500/month, and that about 95% of enterprise AI usage still runs on frontier models (Glean CEO Arvind Jain). On competition, Microsoft, Google (Gemini 3.5 Flash at half to a third the price), and Amazon's in-house chips are all pushing low-cost models. Anthropic declined to comment.

Claude Developer Platform official release note (2026-05-12): Fast mode (research preview) now supports the Claude Opus 4.7 model. By setting `speed: "fast"`, `model: "claude-opus-4-7"`, and the `fast-mode-2026-02-01` beta header in an API request, developers get significantly faster output token generation (up to roughly 2.5x output tokens/sec), priced at 6x standard Opus rates across the full context window (including requests over 200k input tokens). Pricing, rate limits, and access are the same as for Opus 4.6 fast mode; interested customers must join the waitlist. Fast mode does not support the Batch API and is not available for Claude Platform on AWS, Amazon Bedrock, Google Vertex AI, or Microsoft Azure Foundry — it is offered only via the direct Anthropic API (with the beta header) and supported native integrations (Claude Code, Cursor, Windsurf, v0, Warp). Fast mode first launched for Opus 4.6 in February 2026.

Claude Code v2.1.193 released — Added: `autoMode.classifyAllShell` setting to route all Bash/PowerShell commands through the auto-mode classifier instead of only arbitrary-code-execution patterns; auto-mode denial reasons added to the transcript, the denial toast, and `/permissions` recent denials; a `claude_code.assistant_response` OpenTelemetry log event carrying the model's response text (redacted unless `OTEL_LOG_ASSISTANT_RESPONSES=1`); live file-path autocomplete in bash mode (`!`); a startup notice when MCP servers need authentication, pointing at `/mcp`; automatic memory-pressure reaping for idle background shell commands (disable with `CLAUDE_CODE_DISABLE_BG_SHELL_PRESSURE_REAP=1`). Fixes: `/model` and other client-data-gated UI showing stale/empty state right after `/login`; backgrounding (←←) spuriously cancelling with 'N background tasks would be abandoned' when all running tasks carry over to the new session; pinned background agents being re-prompted to 'Continue from where you left off' after every auto-update; backgrounding the main turn spawning a phantom 'general-purpose (resumed)' subagent that re-ran the main conversation; the agent panel hiding sibling agents while viewing a subagent. Improvements: the background-agent launch result no longer instructs Claude to 'end your response' and instead keeps working on other tasks while the agent runs; MCP `headersHelper` auth now re-runs and reconnects automatically when a tool call returns 401/403; plugin auto-rename now follows marketplace `renames` maps automatically and updates settings to the new name; clearer `/add-dir` message when the directory is already a working directory.

TechCrunch: third-party data shows Claude rapidly gaining share in the paid-consumer market dominated by ChatGPT — TechCrunch reported on 2026-06-25 (reporter Julie Bort), citing three third-party data providers, that Anthropic's Claude is rapidly expanding its paid-consumer base: credit-card transaction analyst Indagari (covering ~28 million US consumers, analyzing weekly transactions through 2026-05-10) shows Claude consumer revenue up ~75% since January 2026; online-education platform DataCamp (~20 million users) shows demand for 'Claude' courses surging 18x in the past 30 days, with 'Claude' now its most-searched term, and among self-directed consumers Claude course demand leading ChatGPT 3:1. The report notes ChatGPT still leads by a wide margin in absolute users, revenue, and corporate training, but Claude is narrowing the gap; the growth came despite Anthropic's most capable models Fable 5/Mythos 5 being taken offline by a 6/12 US export-control directive. Anthropic declined to comment on the analysis.

Fable 5 still not back — Anthropic staff debunk the viral 'v2.1.190 lets you pick Fable 5' rumor, confirming 'zero traffic' to the model — After the 6/12 US export-control directive forced Anthropic to suspend Claude Fable 5 and Mythos 5, X posts went viral claiming Claude Code v2.1.190 users could again select Fable 5 in the model picker, even showing weekly usage limits. Anthropic Head of Growth Amol Avasare dismissed the reports as 'categorically false,' judging the visible model listing to be a front-end UI bug rather than actual availability; another staffer, Sam McAllister, said 'We are currently serving exactly 0 traffic to Fable 5.' Selecting Fable 5 actually shows 'Claude Fable 5 is currently unavailable' linking to the 6/12 suspension notice. As of 2026-06-25 (day 13 of the suspension), Fable 5 is not back.

Claude Developer Platform official release note (2026-05-12): Fast mode (research preview) now supports the Claude Opus 4.7 model. By setting `speed: "fast"`, `model: "claude-opus-4-7"`, and the `fast-mode-2026-02-01` beta header in an API request, developers get significantly faster output token generation (up to roughly 2.5x output tokens/sec), priced at 6x standard Opus rates across the full context window (including requests over 200k input tokens). Pricing, rate limits, and access are the same as for Opus 4.6 fast mode; interested customers must join the waitlist. Fast mode does not support the Batch API and is not available for Claude Platform on AWS, Amazon Bedrock, Google Vertex AI, or Microsoft Azure Foundry — it is offered only via the direct Anthropic API (with the beta header) and supported native integrations (Claude Code, Cursor, Windsurf, v0, Warp). Fast mode first launched for Opus 4.6 in February 2026.

Claude adds Trusted Devices for Remote Control on Team and Enterprise plans — Anthropic announced in the Claude release notes (2026-06-25) that admins on Team and Enterprise plans can now require members to verify their device before remotely viewing or steering a local Claude Code session, adding device-level security verification to remote control access.

Claude Code v2.1.191 released — adds `/rewind` support to resume a conversation from before `/clear` was run. Fixes: scroll jumping to the bottom while reading earlier output during a streaming response; background agents resurrecting after being stopped (stopping an agent from the tasks panel is now permanent); `/voice` now explaining the restriction when disabled by org policy instead of a generic 'not available'; `/login` URL truncated when wrapped across lines in Windows Terminal; Cmd+click on links in fullscreen for Ghostty over ssh/tmux; `claude agents` sending builtin slash commands like `/usage` to background sessions as prompt text (now shows a hint); `claude agents` job rows now using the `[Image #N]` placeholder instead of full filesystem paths for pasted images; hooks with comma-separated matchers (e.g. `"Bash,PowerShell"`) silently never firing; `/permissions` recently-denied approvals now persisting on close instead of being discarded; the agent panel jumping a row past the overflow cap; the welcome splash art overflowing the default 80x24 macOS Terminal; managed-settings `forceRemoteSettingsRefresh` now taking effect via MDM/file policy and sending `Cache-Control: no-cache` to block stale proxy responses. Improvements: sandbox network hosts allowed with 'Yes' now remembered for the rest of the session instead of re-prompting; MCP capability discovery (`tools/list`, etc.) retrying transient network errors with backoff; MCP OAuth discovery and token requests retrying once and headless environments going straight to the paste-the-URL prompt; MCP HTTP 404 errors showing the URL and pointing to your config; Vim-mode prompt-history search (NORMAL `/`) hinting how to reach slash commands; streaming CPU usage reduced ~37% by coalescing text updates to 100ms; reduced long-session memory growth from terminal output cache.

Bloomberg/Reuters: Anthropic writes to US senators and the White House, accusing Alibaba of waging the largest distillation attack to date on Claude — nearly 28.8 million exchanges — Bloomberg and Reuters reported on 2026-06-24 that Anthropic, in a 6/10 letter to the Senate Banking Committee (including Senators Tim Scott, Elizabeth Warren, Bill Hagerty, and Andy Kim) and White House officials, accused operators affiliated with Alibaba and its AI lab Alibaba Qwen of running a distillation campaign against Claude between 2026-04-22 and 06-05 — nearly 25,000 fraudulent accounts generating more than 28.8 million exchanges, targeting Claude's most commercially valuable capabilities (software engineering and agentic reasoning). Distillation is the practice of training a smaller, weaker model on a stronger model's outputs to approximate its capabilities at lower cost. Anthropic called it the largest distillation attack to date and the first time it has publicly named a major Chinese tech conglomerate; in February 2026 it had accused three Chinese startups (DeepSeek, Moonshot, MiniMax) of running about 16 million exchanges through roughly 24,000 fake accounts — the Alibaba campaign alone exceeds those combined. An Anthropic spokesperson said "combating the threat of illicit distillation requires coordinated action between government and industry, and we will continue working with Congress and the Administration to maintain American AI leadership," and called on the Trump administration to clarify antitrust guidance so US AI labs can share intelligence, tighten advanced AI chip export controls, and impose penalties on firms using distillation. Alibaba did not immediately respond. Senators Bill Hagerty (R-Tennessee) and Andy Kim (D-New Jersey) plan to introduce a bipartisan amendment to the National Defense Authorization Act (NDAA) later this week to authorize blacklisting or sanctioning Chinese firms found to be improperly accessing US AI model outputs to train competing models. Alibaba's US-listed shares fell about 3% on the day.

Claude Opus 4.8 Fast elevated error-rate incident (resolved) — status.claude.com logged an incident affecting Claude Opus 4.8 Fast: investigation began 2026-06-24 18:22 UTC and resolution was declared at 18:33 UTC, lasting about 11 minutes. It follows the same day's 13:16 UTC Opus 4.8 error-rate incident, continuing the pattern of frontier-model error-rate volatility since Opus 4.8 launched.

TechCrunch: Two of Google's core Gemini researchers, Jonas Adler and Alexander Pritzel, are moving to Anthropic — the fourth senior AI departure in six days — TechCrunch reported on 2026-06-24 that Jonas Adler (who worked on the company's AI coding effort) and Alexander Pritzel (involved in training AI systems), both key contributors to Google's Gemini model, are leaving Google for Anthropic. The report notes the exodus underscores the pressure Google faces from two startups on the cusp of IPOs (OpenAI and Anthropic), where joining before going public offers a rare equity payday. Google has seen a string of top AI talent depart recently: legendary researcher Noam Shazeer (at Google since 2000) left for OpenAI, and 2024 Nobel chemistry laureate and AlphaFold pioneer John Jumper left Google DeepMind for Anthropic.

ABC7 (7 On Your Side): Martinez, California man says a malicious third-party Claude skill racked up ~$315 in unauthorized gift-card charges billed in euros — ABC7 San Francisco reported on 2026-06-24 that Jason Weitzman, a former information-security director, bought an Anthropic subscription to use Claude for his tech company and was flagged by Chase for multiple unauthorized gift purchases billed in euros, two of which (~$315 total) went through, with more attempts over the next two days. Weitzman traced the activity to a third-party 'skill' (add-on) connected to Claude that instructed Claude to use the payment information and digital wallet stored on his computer to attempt to buy various gift accounts; his Anthropic account was banned and he was locked out. After media (7 On Your Side) intervened, Anthropic restored his access, refunded the disputed charges, and added security measures. His bank initially denied the fraud claim, citing the Claude receipt as proof of authorization, before acknowledging the merchant resolved it directly with the customer. The case highlights supply-chain and payment-security risks in Claude's third-party skill/add-on ecosystem.

Claude Code v2.1.190 released — the official release notes list only 'Bug fixes and reliability improvements,' with no named changes; it is a maintenance patch following v2.1.187 on 6/23 (v2.1.188 and v2.1.189 were not publicly released).

Claude Opus 4.8 elevated error-rate incident (resolved) — status.claude.com logged an incident affecting claude.ai, Claude Console (platform.claude.com), Claude API (api.anthropic.com), Claude Code, and Claude Cowork: investigation of elevated Opus 4.8 error rates began 2026-06-24 13:16 UTC and resolution was declared at 14:56 UTC, lasting about 1 hour 40 minutes. It continues the pattern of frontier-model error-rate volatility since Opus 4.8 launched.

Anthropic published a deep-dive on the 'agent identity' access model behind Claude Tag — shifting from individual user permissions to workspace-level identities and giving the AI its own accounts. On 2026-06-24, Anthropic's claude.com blog explained the agent identity access model behind the Claude Tag launched on 6/23: Claude no longer acts on behalf of a single user but holds its own account in each system it touches. Access control shifts from 'what can this user do?' to 'what can this agent do in this compartment?' — a team member without direct access to a repository can still ask Claude to read it if the channel's profile permits. Admins establish baseline Claude identities at the workspace level, defining the tools, repositories, API keys, and skills Claude can access; each channel inherits the default identity and can override it (e.g., granting engineering channels GitHub and warehouse access while restricting CRM connections to specific private channels). Private channels receive distinct identities, public channels share the workspace identity, and what Claude learns in a private channel 'never appears in the wider workspace.' It is an architectural and security follow-up to the 2026-06-23 Claude Tag (beta) launch, addressing the permission problem that 'there's no single choice of person that'd be right all of the time.'

The Next Web: US legal-tech startup Legion LegalTech sues the US government, arguing export controls have no authority to block Anthropic's hosted AI models — The Next Web reported on 2026-06-24 that Legion LegalTech Corp, an AI-litigation startup based in San Jose, California, filed suit that day against the US government in federal court in Washington, DC, challenging the Commerce Department Bureau of Industry and Security (BIS) June 12 directive ordering Anthropic to disable Claude Fable 5 and Mythos 5 for "any foreign national" (inside or outside the US, including the company's own foreign-national staff). Legion argues the directive exceeds the statutory authority granted under export-control law and emergency-powers legislation — existing export controls do not cover hosted AI models or their outputs, and the order is overly broad, effectively blocking access for everyone rather than only foreign nationals. Legion says its Canada-based developers immediately lost model access, causing "immediate, irreparable, and existential" harm to its business, stressing that "competitive ground lost during a suspension cannot be regained." It marks the first time Washington has used export controls to pull a commercial AI product (rather than hardware) off the market, making it a bellwether case testing whether US export-control authority applies to software AI. (Fable 5 was still suspended at the time of filing; Commerce later fully lifted the controls on 6/30 and Fable 5 returned on 7/1.)

Claude Code v2.1.187 released — adds a `sandbox.credentials` setting to block sandboxed commands from reading credential files and secret environment variables, org-configured model restrictions across the model picker, `--model`, `/model`, and `ANTHROPIC_MODEL` (a "restricted by your organization's settings" message appears when a restricted model is selected), and mouse-click support for selection menus (permission prompts, `/model`, `/config`, etc.) in fullscreen mode. Fixes: `--resume` failing with "No conversation found" when the original `-p` run produced no model turns; `--json-schema` and workflow `agent({schema})` structured output no longer re-calling StructuredOutput indefinitely after success (follow-up turns now reliably return structured output); remote MCP tool calls that hung for 5 minutes now abort with an error (override via `CLAUDE_CODE_MCP_TOOL_IDLE_TIMEOUT`); Claude Code Remote sessions starting ~2.7s slower after the agent-proxy CA system-trust install; pasted Korean/CJK text turning to mojibake in terminals delivering paste as per-byte extended-key events; `/update` over Remote Control hanging on a startup trust dialog; background jobs stuck in "working"; channel connections dropping after the agents view and after `/bg`/`/tui`/`/update`; agent stop notifications mis-attributing who stopped the agent (wording now "finished"/"stopped"); subagent depth tracking for resumed and forked subagents; leaked agent worktree registrations; Cmd+click not opening URLs in Ghostty fullscreen on macOS; `claude --help` omitting `--bg`/`--background`; Esc/Ctrl-C/Ctrl-D not working while `/share` uploads; and the VSCode extension freezing when resuming a large session. Improvements: `/install-github-app` workflow setup is now optional, `/btw` gains ←/→ navigation through earlier answers, and `/plugin` surfaces recently unused plugins for cleanup.

Claude.ai elevated error rate incident (resolved) — status.claude.com logged an incident affecting only claude.ai (titled "Claude.ai is experiencing elevated error rates," with no specific model named): investigation began 2026-06-23 18:24 UTC, a fix was implemented and monitoring began at 18:32 UTC, with resolution declared at the same time, lasting about 8 minutes. It follows the same day's 14:19 UTC multi-model error-rate incident, continuing the pattern of recurring frontier-model error rates since Opus 4.8's launch.

Elevated error rate across multiple Claude models (resolved) — status.claude.com logged an incident affecting claude.ai, Claude Console (platform.claude.com), Claude API (api.anthropic.com), Claude Code, and Claude Cowork (titled 'elevated error rate across multiple models'; no specific model named): investigation began 14:19 UTC on 2026-06-23, the issue was identified and a fix implemented by 14:25 UTC, the fix was deployed and monitoring began at 14:53 UTC, error rates returned to normal at 15:33 UTC (impact window 14:08–15:33 UTC, ~1h25m), and the incident was formally resolved at 16:44 UTC. External monitor DownDetector saw over 8,000 outage reports for Claude on 6/23, adding to a recent run of error-rate incidents and continuing the pattern of frontier-model error-rate volatility since Opus 4.8 launched.

Fortune: Anthropic engineering head Fiona Fung says heavy agentic-AI use has made employees' work a 'lonely experience,' hinting at a broader Big Tech morale issue — Fortune reported on 2026-06-23 that Fiona Fung, the engineering lead for Anthropic's Claude Code and Cowork teams, said on a recent episode of Lenny's Podcast (released 2026-06-21) that agentic-AI usage in the workplace had grown so much it was making people's day-to-day more solitary: 'after a while, we felt it could start being a lonely experience because we all started just working with our agents so much.' Anthropic responded by introducing hackathons and pair-programming lunches to restore in-person collaboration; the piece quotes one Anthropic employee saying, 'On days where everything works well, I can't help but think nothing I do matters, everything is automated and better and faster than I ever will be.' Before Anthropic, Fung spent 11 years at Microsoft (Visual Studio and TypeScript), then moved to Meta, where she launched Facebook Marketplace (now >$100B GMV annually), worked on Meta's first smart and AR glasses, and led infrastructure, growth, integrity, and safety teams at Instagram.

Legion sues the US government over the block on Claude Fable 5 access — the legal-tech startup says the export-control directive inflicts "immediate, irreparable and existential" harm on its operations — multiple outlets (Bloomberg, Gizmodo) reported on 2026-06-23 that Legion, a startup building tools for lawyers, filed suit Tuesday in federal court in Washington against the US government, challenging the national-security-based export-control directive that requires Anthropic to suspend access to Claude Fable 5 and Mythos 5 for all foreign nationals (inside or outside the US, including foreign-national Anthropic employees). Legion is US-based but its software-development team includes Canadian nationals working from Canada who are affected by the directive. Legion argues Fable 5 is critical to its business, saying "each day the directive remains in force disrupts Legion's product and operations, sidelines its engineers, and erodes the company's ability to survive in a field defined by continuous access to the most capable models." This is the first suit by a downstream customer against the government following the 6/12 export-control order under which Anthropic suspended Fable 5/Mythos 5 for all customers, extending the legal dispute over applying export-control law to AI model "usage."

Bloomberg/TechCrunch: Anthropic investor Menlo Ventures raises its largest-ever $3B fund; its Anthropic stake is now valued near $14B — Bloomberg and TechCrunch reported on Tuesday 2026-06-23 that Menlo Ventures announced $3 billion in fresh capital — its largest raise ever — on its 50th anniversary, split across two flagship funds: Menlo Ventures XVII (seed to Series A) and Menlo Inflection IV (Series B and growth), targeting AI infrastructure, frontier models, and AI-native applications across enterprise, healthcare, and consumer markets. Menlo has invested roughly $1B in Anthropic over the years, plus about $500M across multiple follow-on rounds; according to people familiar, its stake is now worth nearly $14B as Anthropic's valuation has soared past $900B and approaches the trillion mark. Managing partner Shawn Carolan calls the Anthropic investment a 'bet-the-firm moment'; Menlo and Anthropic also co-launched the $100M Anthology fund in 2024 to back seed-to-expansion-stage AI startups.

SCMP: Anthropic's global Fable 5 shutdown opens market space for China's Zhipu AI open-weight model GLM-5.2 — The South China Morning Post reported on 2026-06-23 that, under the US government's June 12 export-control directive, Anthropic chose to disable Claude Fable 5 globally on June 12 (rather than implement nationality-based access controls), prompting developers and businesses worldwide to reassess their reliance on US AI technology. Chinese startup Zhipu AI seized the moment, launching its open-weight model GLM-5.2 just one day after Anthropic suspended Fable 5; Zhipu AI founder Tang Jie called the restrictions "deeply regrettable" and emphasized GLM-5.2's "fully open" nature. The report notes US frontier models retain a significant technological lead, but with Fable 5/Mythos 5 offline under export controls, an opening is emerging for Chinese open-weight models to gain market share.

Anthropic launches Claude Tag (beta) — @Claude joins Slack channels as a team member that can run async tasks spanning hours or days — Anthropic announced Claude Tag in its newsroom on 2026-06-23: teams can @Claude in a Slack channel to delegate tasks while they focus on other work. Features include multiplayer use (one Claude per channel interacting with everyone), contextual learning from channel history and connected data sources, proactive flagging of relevant info when ambient behavior is enabled, and independent async execution spanning hours to days. Setup is four steps (pair with workspace, grant tool access, set a monthly spend limit, test); Claude Tag replaces the existing Claude in Slack app, with admins configuring per-channel tool access and information permissions. It runs on Opus 4.8, and Anthropic says 65% of its product team's code is written by an internal version of Claude Tag. Rolling out in beta to Claude Enterprise and Team customers with an introductory launch credit for eligible orgs; expansion beyond Slack is planned.

Elevated errors for Claude Opus 4.8 (resolved) — status.claude.com logged an incident affecting claude.ai, Claude Console (platform.claude.com), Claude API (api.anthropic.com), Claude Code, and Claude Cowork: investigation of elevated Opus 4.8 error rates began at 06:28 UTC on 2026-06-23, a fix was implemented and monitoring began at 07:47 UTC, and the incident was declared resolved at 08:45 UTC, lasting about 2 hours 17 minutes. It follows three error-rate incidents on 6/22 (00:37 five models, 08:11 Opus 4.8, 19:14 many models), continuing the recurring frontier-model error-rate pattern since Opus 4.8's launch.

Claude Code v2.1.186 released — Added `claude mcp login <name>` / `claude mcp logout <name>` to authenticate MCP servers from the CLI without opening the interactive `/mcp` menu (with `--no-browser` stdin redirect to complete over SSH), status filtering (press f) in the `/workflows` agent detail view, a Skills section in the `/plugin` Installed tab, a `teammateMode: "iterm2"` setting (warns when auto mode cannot find the it2 CLI), a "Claude Platform on AWS - refresh credentials" option in `/login` when `awsAuthRefresh` is configured, and `!` bash commands now auto-trigger a Claude response (set `respondToBashCommands: false` in settings.json to keep context-only behavior). Changed: `CLAUDE_CODE_MAX_RETRIES` now caps at 15 (use `CLAUDE_CODE_RETRY_WATCHDOG` for unattended sessions); background subagent permission prompts now surface in the main session showing which agent is asking (no longer auto-denied, Esc denies just that tool); `/review <pr>` now uses the same engine as `/code-review medium`. Fixed: streaming requests failing with "Content block not found" or JSON parse errors after waking from sleep, subagent transcript scroll bleeding into the main transcript on exit, `Agent(type)` deny rules and `Agent(x,y)` allowed-types not enforced for named subagent spawns, Esc and Ctrl+C unresponsive while background agents still run after the main turn, missing session cost for usage-based Enterprise and Team subscribers, and workflow `agent({schema})` subagents looping forever on repeated schema validation failures (now aborts after 5 attempts). Improved: `claude mcp get`/`remove` suggest the closest server name on a typo, Memory reminds the agent to compact its `MEMORY.md` index near the size limit, and Skill frontmatter keys accept kebab-case / snake_case / camelCase.

Elevated error rates across multiple Claude models (resolved) — status.claude.com logged an incident affecting claude.ai, the Claude API (api.anthropic.com), Claude Code, and Claude Cowork (titled "elevated errors across many models"; the status page named no specific model): investigation began at 19:14 UTC on 2026-06-22, a fix was implemented and monitored at 19:20 UTC, and it was resolved at 19:45 UTC, lasting about 31 minutes. It was the day's third error-rate incident after the 00:37 UTC five-model and 08:11 UTC Opus 4.8 events, continuing the pattern of recurring frontier-model error rates since Opus 4.8's launch.

TechCrunch: Anthropic updates its privacy policy — Claude may ask you to upload a government-issued ID to verify identity and age (effective 2026-07-08). Reported by TechCrunch on 2026-06-22, Anthropic's updated privacy policy (effective 2026-07-08) lets it ask users, "in certain circumstances," to upload government documents (passport or driver's license) to verify identity and age. It targets a small subset of consumer-plan users (Free, Pro, Max) whose accounts are flagged for potentially fraudulent activity, serving as a "verify rather than ban outright" appeals path; it does not affect Team, Enterprise, or Platform (Commercial Terms) users. Data collected includes a scanned ID image and its details (ID number, date of birth), a selfie photo or video, a facial-geometry template (considered biometric data in some jurisdictions such as Illinois), and verification results. Anthropic uses Persona for verification, says the data is not used to train models and is used only to confirm identity and meet legal and safety obligations, and cites age-verification requirements across various states and countries as the rationale.

Micron and Anthropic announce a strategic agreement spanning memory/storage AI architecture, supply, enterprise Claude adoption, and a Series H investment — Micron Technology issued a press release on 2026-06-22 announcing a strategic agreement with Anthropic covering memory and storage AI architecture design, supply-and-demand collaboration, enterprise-wide adoption of Claude across Micron, and a strategic investment in Anthropic's $65B Series H round (which closed in late May at a $965B post-money valuation). The deal makes Micron Anthropic's primary memory and storage supplier; Micron's HBM (high-bandwidth memory), DRAM, and SSD products underpin performance, power efficiency, and total cost of ownership (TCO) across AI training and inference. Claude is already deployed across Micron's engineering, manufacturing, and enterprise functions to accelerate coding and enable more advanced agentic use cases.

Elevated error rates for Claude Opus 4.8 (resolved) — status.claude.com logged an incident affecting claude.ai, the Claude API (api.anthropic.com), Claude Code, and Claude Cowork: investigation of elevated Opus 4.8 error rates began 2026-06-22 08:11 UTC, continued at 08:12 UTC, a fix was implemented and entered monitoring at 09:59 UTC, and the incident was declared resolved at 14:44 UTC — about 6 hours 33 minutes in total. It follows the same day's 00:37 UTC five-model error-rate incident, continuing the pattern of recurring frontier-model error rates since Opus 4.8's launch.

Claude Desktop's full experience comes to AWS, Google Cloud, and Microsoft Foundry — Chat, Cowork, and Claude Code unified in a single app. On 2026-06-22, Anthropic's claude.com blog announced that organizations deploying Claude Desktop through AWS, Google Cloud, and Microsoft Foundry now get the complete feature set in one application (previously limited to Claude Cowork and Claude Code), adding Chat for quick answers and problem-solving. Inference runs on the enterprise's own cloud in the regions they configure, and conversation history is stored locally. Enterprise deployment controls include SSO integration with existing identity providers (IAM Identity Center, Microsoft Entra ID, Okta), policy templates compatible with Intune/GPO/Jamf, pre-rollout validation, granular access control with separate policy keys per surface, and a local Microsoft 365 connector option for strict data-residency requirements.

Elevated error rates across multiple Claude models (largely resolved) — status.claude.com logged an incident affecting claude.ai, the Claude API (api.anthropic.com), Claude Code, and Claude Cowork, naming elevated error rates on five models: Opus 4.8, Opus 4.7, Opus 4.6, Sonnet 4.6, and Haiku 4.5. Investigation started 2026-06-22 00:37 UTC, the issue was identified and a fix deployed by 01:11 UTC, and Opus 4.8 was reported recovered at 01:16 UTC — roughly 39 minutes from investigation to recovery. It corresponds to the surge of Claude outage reports on DownDetector around 8:31 PM US Eastern on June 21 (00:31 UTC), continuing the pattern of recurring frontier-model error rates since Opus 4.8's launch.