🗞️ AI Daily Briefing — 2026-06-24

🔥 Top Story

Anthropic launches Claude Tag — an always-on AI teammate that lives in your Slack channels. Announced yesterday, Claude Tag (@Claude) is a shared, persistent AI identity that any team member can summon, hand tasks to, and pick up mid-conversation. It breaks requests into stages, works through them autonomously, and — with “ambient” mode enabled — proactively flags forgotten threads and surfaces relevant info across channels. Anthropic says 65% of its own product team’s code is written by their internal version. Available now in research preview for Claude Enterprise and Team customers. (Anthropic) (TechCrunch) (Bloomberg)

🚀 Model & Research News

  • Google DeepMind’s brain drain sends GOOGL tumbling 7%. In the span of five days, Google lost Nobel laureate John Jumper (AlphaFold) to Anthropic and Transformer co-inventor Noam Shazeer to OpenAI. Alphabet had its worst single-day drop in over a year on Monday. Analysts are questioning whether Google’s internal bureaucracy is pushing top talent toward the pre-IPO upside at rivals. (Axios) (CNBC) (Fortune)
  • Fable 5 / Mythos 5 still offline — day 12 of the US export ban. No restoration timeline. Anthropic’s status page shows both models remain suspended following the June 12 government directive citing a narrow Fable 5 jailbreak. Anthropic disagrees with the standard, saying it would “essentially halt all new model deployments for all frontier model providers.” Claude Opus 4.8 remains unaffected and holds #1 on the Arena leaderboard. (Anthropic) (Fortune)
  • Gemini 3.5 Pro still MIA with 6 days left on Google’s June deadline. Prediction markets sit at roughly 50-55% for a public launch by June 30. The model is enterprise-only on Vertex AI. Sundar Pichai’s I/O promise of “give us until next month” is looking increasingly strained. (TechCrunch) (Swisher Post)
  • Arena leaderboard (6.9M votes, 367 models): Claude Opus 4.8 holds #1 at ~1510 ELO, followed by GPT-5.5 Pro, Gemini 3.1 Pro Preview, Claude Opus 4.7, and GPT-5.5. Top five clustered within ~55 ELO points — tightest spread on record. (Arena.ai)

🛠️ Tools & Developer Updates

  • Claude Tag redefines “AI coworker.” Beyond the headline product, the enterprise features are notable: admins can scope tools, memories, and token budgets per channel; Claude learns org context over time so teams don’t re-explain from scratch. Auto-migration for existing Slack integrations happens August 3. (Anthropic) (Neowin)
  • “Tokenmaxxing” officially dead. The enterprise AI spending backlash continues: Uber burned its entire 2026 AI budget in four months (largely on Claude Code), imposed a $1,500/month/employee cap. Microsoft cancelled Claude Code licenses for several product divisions. Meta killed its internal token leaderboard. Uber’s COO: “That link [between token usage and consumer innovation] is not there yet.” (Fortune) (TechCrunch)

💰 Funding & Business

  • The AI IPO supercycle is here. Anthropic filed its S-1 confidentially on June 1 (post-$65B Series H at $965B valuation). OpenAI filed June 8, targeting up to $1T at IPO. SpaceX began trading June 12 after the largest IPO on record ($75B raised). These three could be the largest IPOs ever recorded. (CNBC) (CNBC)
  • Anthropic paying SpaceX $1.25B/month for Colossus compute. The deal gives Anthropic access to ~325,000 NVIDIA GPUs across Colossus I and II through May 2029 — up to $45B over the full term. Grok’s decline left SpaceX’s data centers at ~11% utilization; Anthropic fills the gap. (Axios)
  • OpenAI generating $2B/month in revenue but forecasting a $14B loss for 2026 (adjusted operating margin: -122%). Enterprise now 40%+ of revenue. Sam Altman disclosed that Anthropic has overtaken OpenAI in corporate AI spending among enterprise customers — a first. (CNBC)

🐦 Notable from the Timeline

  • @sama welcomed Noam Shazeer to OpenAI: “noam is one of the people I have most wanted to work with since the very beginning of openai. only took 10 years.” Shazeer joins as Lead for Architecture Research.
  • @alexandr_wang — one year into his role as Meta’s Chief AI Officer (after Meta’s $14.3B investment in Scale AI), CNBC reports the Muse Spark model landed but Meta’s AI products still “struggle for differentiation.” Google and OpenAI are both reportedly severing deals with Scale AI over the Meta arrangement. (CNBC)
  • Andrej Karpathy is now at Anthropic, leading a new team using Claude to accelerate pretraining research — a major talent coup following his stints at OpenAI, Tesla, and Eureka Labs. (TechCrunch)
  • Enterprise AI reality check: 56% of CEOs report no revenue or cost benefit from AI spending. Average enterprise AI spend hit $11.6M in 2026, with costs reaching $7,500/month per AI-heavy worker. (Fortune)

📊 Benchmark Watch

Arena leaderboard holds steady with Claude Opus 4.8 at #1. The big question is whether Gemini 3.5 Pro will shake things up if Google makes its June deadline — the model’s 2M-token context window and “Deep Think” reasoning mode could be significant. Fable 5’s brief #1 reign before the export ban suspension means we still don’t have a stable public reading of its true ELO. (Arena.ai)

🔗 Worth Reading

  • “Google DeepMind loses star power” — Axios deep dive on the back-to-back departures and what they signal about Google’s ability to compete for talent against pre-IPO rivals
  • “Tokenmaxxing is over” — Fortune’s postmortem on why “use as much AI as possible” was a flawed strategy and the enterprise ROI reckoning that followed
  • Introducing Claude Tag — Anthropic’s official blog post detailing the architecture, enterprise controls, and ambient behavior features of their new Slack-native AI teammate