April 17, 2026 — Your daily AI briefing
Three forces converged in the AI world today: a landmark annual report declared the US-China race effectively tied, Anthropic quietly doubled its million-dollar customer base while previewing a model that can find zero-days at scale, and Meta finally showed what it bought when it hired Alexandr Wang for $14 billion.
Stanford's AI Index Verdict: The Lead Is Gone
The 2026 Stanford AI Index — the field's most cited annual scorecard — landed today with a finding that should rattle policymakers: China has "nearly erased" the United States' advantage in frontier AI. On the Arena community leaderboard, which ranks models by human preference, the two nations are now effectively neck and neck. The best models on both sides, including Anthropic's Claude Opus 4.6 and Google's Gemini 3.1 Pro, now crack 50% on Humanity's Last Exam — a benchmark that was considered essentially unsolvable just 18 months ago.
The report also quantifies AI's energy appetite for the first time at scale: global AI data centers now draw 29.6 gigawatts of power — enough to run the entire state of New York at peak demand. Meanwhile, the adoption picture remains paradoxical: 80% of enterprise workers are actively avoiding or rejecting AI tools, even as 60% of consumers say they'd trust AI more if it could explain its reasoning. The gap between AI capability and AI trust has never been wider. (IEEE Spectrum, Fortune)
Anthropic's Mythos Preview: Zero-Days at Scale
Anthropic had a quietly explosive week. The company disclosed that its Claude Mythos Preview — the unreleased successor to its current flagship — scored 93.9% on SWE-bench Verified and 94.6% on GPQA Diamond, putting it at or above every publicly available model on both coding and graduate-level science reasoning. More striking: the model reportedly identified thousands of zero-day vulnerabilities across every major operating system and browser during internal testing.
On the business side, Anthropic revealed it has more than doubled the number of customers paying $1 million or more annually since February — a signal that enterprises are moving from pilots to production at pace. Meanwhile, its production Mythos 5 model, built on 10 trillion parameters, is already in deployment for advanced cybersecurity and coding workloads. (Applying AI, Next Platform)
Meta's Muse Spark: The $14B Bet Makes Its Debut
Nine months after Meta spent an estimated $14 billion to bring Scale AI's Alexandr Wang in-house, the company's new Meta Superintelligence Labs has finally shipped its first model. Dubbed Muse Spark (internally code-named "Avocado"), it's the inaugural release in Meta's new Muse series. Full capability details remain sparse, but the launch signals that Meta's costly talent acquisition is beginning to produce tangible output — and that the company intends to compete seriously at the frontier rather than ceding ground to Anthropic, Google, and OpenAI. (CNBC)
What to Watch
The Stanford AI Index's energy data will likely accelerate regulatory scrutiny of data center buildouts — expect EU and US grid agencies to weigh in within weeks. On the model front, Anthropic's full Mythos release (not just the preview) is the imminent wildcard: if the zero-day vulnerability findings hold up under independent review, the security community will have a lot to process. And Meta's Muse Spark needs a public benchmark outing — until then, the $14B question remains unanswered.