Anthropic News
Podcast 🎧 • Video 📽
Compute & Chips
AMD Invests Up to $5B in Anthropic — 2 Gigawatts of MI450 GPUs Starting First Half of 2027
Advanced Micro Devices announced Wednesday a strategic partnership with Anthropic under which AMD will invest up to $5 billion in Anthropic and Anthropic will deploy up to 2 gigawatts of AMD Instinct MI450 Series GPUs in AMD Helios rack-scale solutions. The $5 billion equity commitment is milestone-contingent and phased over time. The first gigawatt of GPU capacity is scheduled for deployment in the first half of 2027. Anthropic will acquire some chips outright for its own data centers and lease additional capacity through cloud providers and neocloud operators. AMD and Anthropic will also launch a multi-year engineering collaboration, with AMD integrating Claude across its software and product development. AMD CEO Lisa Su called it a deployment at gigawatt scale. The deal adds AMD to an infrastructure partner list that already includes Google, Amazon, SpaceX Colossus 1, Broadcom, and TeraWulf, and comes alongside preliminary talks with Meta reported last week. Anthropic acknowledged in April that surging Claude demand had created strain on its infrastructure, affecting reliability during peak hours.
Sources: Reuters · Bloomberg · CNBC · Wall Street Journal · Quartz
Politics & Money
Anthropic Donates Second $20M to Public First Action — Total AI Safety Super PAC Commitment Reaches $40M
Anthropic announced Wednesday a second $20 million donation to Public First Action, the political nonprofit that advocates for mitigating major AI risks, bringing the company's total commitment to the organization to $40 million. The first $20 million was donated in February. The announcement lands the same week that FEC filings revealed Anthropic CEO Dario Amodei personally contributed $1 million to Public First's affiliated super PAC in May, and that five other Anthropic employees added a combined $1.26 million. Public First Action backs political candidates who support stronger AI regulation. The AI industry's combined super PAC commitments ahead of the 2026 midterms have now crossed $265 million.
Sources: Reuters · WTVB
Anthropic Q2 Lobbying Hits Record $1.97M — Surpasses Nvidia, Driven by Commerce Department Export Control Fight
Federal lobbying disclosures released Tuesday show Anthropic spent $1.97 million in Q2 2026, a 26% jump from Q1 and the company's highest quarterly total since it began lobbying in March 2024. The figure tops Nvidia's Q2 spend and brings Anthropic's first-half 2026 total above $3.5 million — already exceeding its full-year 2025 spend of $3.1 million. The surge traces directly to the Commerce Department's June shutdown of Claude Fable 5 and Mythos 5 over national security concerns, after which Anthropic deployed an expanded in-house team alongside nine outside K Street firms. Lobbying focused on export controls, cybersecurity, AI safety standards, and defense procurement. Anthropic's spend was nearly matched by OpenAI's record $1.2 million. Together the two labs spent $3.17 million in Q2, up 23% from Q1.
Sources: Axios · CNBC · The Next Web · Cryptopolitan
Claude Product
Claude Code Ships Emoji Autocomplete, Memory Leak Fix, and Tighter Subagent Controls
Anthropic released a Claude Code update adding emoji shortcode autocomplete — type :heart: to insert ❤️ or partial strings for suggestions, with an opt-out setting — along with clearer warnings when transcript writes fail due to a full disk or disabled session saving, previously a silent data-loss condition. The update fixes a memory leak where truncated MCP tool outputs retained full untruncated results in memory for the duration of a session. It also tightens controls for subagents, budget management, and background sessions, and improves Windows stability, accessibility, and resume flows.
Sources: Releasebot · Anthropic
Project Glasswing
90-Day Report Still Pending — 14 Days to Black Hat, AMD Deal Expands the Infrastructure Case for Mythos at Scale
Project Glasswing's 90-day progress report has not been published as of July 22. The program operates with approximately 200 partner organizations across 15 countries scanning critical infrastructure codebases with Claude Mythos 5. Black Hat USA opens August 5 in Las Vegas — 14 days away. Today's AMD deal materially strengthens the infrastructure argument for expanding Mythos access: with 2 gigawatts of dedicated GPU capacity coming online in 2027, the compute constraint that has limited Glasswing partner access becomes more addressable. Anthropic has made no announcement on the report.
Sources: Anthropic · Reuters · Decryption Digest
Claude Mythos
Mythos 5 Access Unchanged — AMD Deal and Google's Glasswing-Competing Gemini 3.5 Flash Cyber Change the Landscape
Claude Mythos 5 remains restricted to vetted Glasswing partners as of July 22, unchanged since the July 1 restoration. Two developments today add context. First, the AMD infrastructure deal begins addressing the compute scarcity that has constrained Mythos deployment. Second, Google released Gemini 3.5 Flash Cyber on July 21 — a security-tuned model restricted to governments and trusted partners — placing it in direct competition with Mythos in the AI cybersecurity category alongside Microsoft's Project Perception. General availability for Mythos-class models carries no confirmed timeline.
Sources: Anthropic · Reuters · Build Fast With AI