Curated weekly
What changed, and what it changes for you. Every item is triaged: act, verdict change, watch, or safe to ignore.
44 Important · 24 verdict changes · 31 FYI · 1 watching · 100 total
ImportantPolicy & Regulation ·
The European Commission gains fining power under the EU AI Act tomorrow, and OpenAI arrives with a documented copyright compliance gap. The first enforceable AI law with real penalties has its first clear test case.
Read the full callImportantPolicy & Regulation ·
Sam Altman endorsed deliberate AI pacing for the first time after GPT-5.6 Sol's Hugging Face breach (calling it "the first security incident I have felt very viscerally") as 1,100+ frontier-lab employees petition Washington three days before the White House AI Review deadline.
ImportantModels ·
MAI-Cyber-1-Flash scores 96% on CyberGym (12 points above Mythos 5 at half the cost) alongside Project Perception, an agentic security platform launching Aug 3.
ImportantModels ·
OpenAI cut GPT-5.6 Luna 80% and Terra 20%. Sol autonomously rewrote production GPU kernels, the first frontier model to self-optimize its serving costs and pass the savings to customers.
ImportantIndustry ·
Azure passed $100B in annual revenue, a Copilot super app is coming this year, and Microsoft's own MAI models signal a decoupling from OpenAI and Anthropic. The enterprise AI inflection point has arrived.
ImportantPolicy & Regulation ·
Sam Altman demoed OpenAI's next model to Congress, then backed AI pacing at the White House, two days before the August 1 deadline.
Foundation and frontier model releases, updates, and capability changes.
Important
Anthropic shipped Claude Opus 5 on July 24 at $5/$25 per 1M tokens, near-Fable 5 intelligence at half the price, with SOTA Frontier-Bench scores and 85% fewer safety interruptions. Our verdict: Recommended.
Important
Google shipped three efficiency-focused Gemini Flash models, while the flagship 3.5 Pro misses its June target and Gemini 4 pretraining begins.
D
Important
Poolside's 118B MoE open-weight model beats DeepSeek-V4-Pro-Max and Inkling on Terminal-Bench 2.1 at 8B active params. $0.10/1M tokens, runs on a single DGX Spark.
Important
Anthropic reversed course: Max/Team Premium plans now include Fable 5 at 50% limits starting July 20, the fourth access-policy change in four weeks. Pro and Team Standard get a one-time $100 credit.
+1Important
Moonshot AI's Kimi K3 is a 2.8T open-weight model that beat Fable 5 on Arena Code WebDev at $3/$15, the third Chinese frontier release this month.
Government action, regulation, and AI governance developments.
OImportant
Sam Altman proposed donating ~5% of OpenAI (worth $42.6B at its $852B valuation) to a US sovereign wealth fund. The first AI company to voluntarily put equity on the table for public ownership.
Important
The US lifted its ban on Anthropic's Fable 5, restoring the model developers called the best coding model in the world. It is the first government AI shutdown undone.
Important
Gov. Newsom signed a deal giving all California agencies 50% off Claude access. The Pentagon's "supply-chain risk" label "just didn't come up," and a blue-state procurement template is born.
Important
Mythos 5 access restored for 100+ US orgs on June 26, the first crack in the June 12 export ban. Non-American employees included. Fable 5 stays blocked.
GImportant
The Trump administration asked OpenAI to stagger GPT-5.6 behind case-by-case government approval. Altman confirmed June 25. The regulatory divide is now public record.
AI coding assistants, editors, and developer-facing tooling.
Important
Cursor dumped usage-based billing for five fixed-price tiers. Predictable monthly costs replace unpredictable overages. Pick your tier and stop watching the meter.
CImportant
Cursor announced an in-house AI model, AI-native Git, and a mobile monitoring app: three moves that make it the first vertically integrated AI coding platform.
+1Verdict changed
Z.ai launched ZCode, a free agentic coding desktop app powered by GLM-5.2, with plans starting at $16.20/mo, prices that undercut Cursor and Claude Code by wide margins. The MIT-licensed model behind it eliminates the regulatory kill-switch risk US tools now carry.
Verdict changed
Microsoft's MAI-Code-1-Flash hits GA on GitHub Copilot: its first in-house coding model, cutting OpenAI dependency and reshaping the AI coding tools market.
Verdict changed
Claude Code creator Boris Cherny, who declared AI 'solved coding' in January, now admits 100% AI-generated code is 'problematic', shifting the industry from hype to cost realism.
Security incidents, vulnerabilities, and safety disclosures.
Important
On June 10, 2026, Anthropic formally accused Alibaba of running 25,000 fraudulent accounts to extract Claude's capabilities, the largest known AI model theft in history. The US-China AI cold war just escalated.
Verdict changed
Zhipu AI's open-weight GLM-5.2 matches Anthropic's restricted Mythos 5 on cybersecurity benchmarks. MIT-licensed, self-hostable on consumer hardware: the banned capability is now freely downloadable.
Verdict changed
The Mythos 5 export ban is fragmenting the AI security market, and Asian startups 360 Security and Sakana AI are the first to capitalize with ban-proof alternatives.
+1FYI
FAR.AI's new leaderboard found Grok 4.5 jailbreakable for $58 and Gemini 3.1 Pro for $278. Claude Fable 5 and GPT-5.6 Sol withstood every attack: a hundredfold gap now public.
FYI
China's NVDB labeled Claude Code a security backdoor, escalating the Alibaba ban to a state-level warning. Anthropic says the tracking code is being removed.
Mergers, acquisitions, milestones, and broader industry moves.
OImportant
Apple sued OpenAI for hardware trade secret theft, naming 400+ ex-Apple staffers. The real target: OpenAI's still-unlaunched consumer device.
Important
Bloomberg reports Microsoft is routing Office AI to MAI models instead of OpenAI and Anthropic, and Suleyman aims to eliminate external costs.
FYI
On July 27, Cognizant became a Global Premier Claude Partner embedding the model across manufacturing, healthcare, and insurance. Stock jumped 8%. The next day, ICON signed a multi-year deal to deploy Claude across 40,200 clinical trial staff. Together they confirm Anthropic's channel-first enterprise strategy.
FYI
Nadella publicly called Fable 5 'editorially controlled' to employees, exposing the growing conflict between Microsoft's AI ambitions and its $30B Anthropic partnership.
FYI
Anthropic confidentially filed its S-1 in June, putting the $965B AI lab on track for an October Nasdaq IPO ahead of OpenAI, CNBC and Bloomberg reported.
Compute, hosting, and the infrastructure layer underneath AI.
Important
OpenAI and Broadcom unveiled Jalapeño, a custom LLM inference chip built in nine months, the opening shot of a silicon war that will reshape AI economics.
AVerdict changed
The BIS warns AI infrastructure overinvestment, circular financing, and off-balance-sheet leverage could trigger a 2008-scale crisis, as AI stocks shed $2.7T this June.
Verdict changed
Google told Meta it can't supply the full Gemini capacity Meta wanted, the first sign frontier AI is supply-constrained even for the largest companies.
MFYI
Anthropic shipped MCP 2026-07-28: a stateless rewrite that eliminates sticky sessions, adds header-based routing, and formalizes extensions. AWS, Microsoft, and Cloudflare are already in production with the new spec.
Pricing, plan, and cost changes across tools and models.
Verdict changed
AI by Zapier moved from flat task-based pricing to three model tiers on June 15. The default is Advanced at 3x the base cost, and tool calls multiply on top.
FYI
Cursor collapsed its pricing from six tiers to four, dropping Individual to $16/mo and absorbing Pro+ ($60) and Ultra ($200). Grok model access is now included at the entry-level paid tier, the first visible integration since SpaceX's $60B acquisition.
FYI
Make's credit pricing at $12–38/mo annual replaces per-operation billing. The August 2025 switch signals how automation platforms charge more for AI without raising base prices.
FYI
n8n expanded Starter AI credits from 50 to 2,300/month and Pro from 150 to 13,700, a 46–91x jump making AI workflow building practical on every plan.
Automation platforms, orchestration, and workflow tooling.
FYI
Zapier's app directory reaches 9,000+ integrations, a 3x lead over Make and 20x over n8n. The moat is real, but per-task pricing still punishes volume.
Watching
n8n's pricing restructure is a win: the new self-serve Business tier (€667/mo) brings SSO, Git, and environments without an Enterprise sales call.
New product and tool launches entering the market.
Autonomous agents, agent frameworks, and agentic runtimes.
Never need to catch up again
The weekly delta: only verdict changes and act-now items. No digest filler.