Anthropic made Fable 5.1 generally available on 1 September at $10 and $50 per million — unchanged from Fable 5. Self-reported Terminal-Bench-Science went from 24.7 to 52.6, cache reads dropped to $0.25, and Mythos 5.1 ships as the trusted-access twin.
Large Language Models03 Sep 20266 min read
Sony Music Publishing and Warner Chappell filed a 48-page complaint in Northern California naming Anthropic, Dario Amodei and Benjamin Mann. They seek up to $150,000 per song across tens of thousands of compositions. Anthropic calls it recycled. The reason it is not routine is a judgment already on the books.
🔥 TrendingNews02 Sep 20268 min read
The Claude Code weekly boost does not expire on 31 August. It runs to 13 September, and on the 14th a permanent level takes effect at 25 percent above the old baseline — which is 17 percent below what users have today. Anthropic published the first framing, deleted it, and reposted with the reduction stated plainly.
🔥 TrendingCode Tools30 Aug 20267 min read
Judge Rita Lin found the Defense Department retaliated against Anthropic for refusing to permit domestic surveillance and autonomous weapons use, violating the First and Fifth Amendments. The government was ordered to rescind all directives. A second case is still live in D.C., where an appeals court ruled the other way in April.
🔥 TrendingNews29 Aug 20268 min read
Anthropic's CEO made the claim on X in mid-August, backed by a reported $400 million biotech acquisition and lab-validated protein results. Cancer researchers responded that cancer is not one disease but more than 200, and that telling patients otherwise is cruel. Both positions deserve to be read.
News28 Aug 20268 min read
Anthropic shipped a Chromium browser inside Claude Cowork on 26 August, rolling out to all paid plans over the following week. It joins the Claude Code browser from July and the Chrome extension, and the difference between them comes down to one thing: whether Claude uses your logins or a clean profile.
AI Agents27 Aug 20267 min read
Anthropic merged the two memory systems on 25 August. Hand Cowork a task and it starts from what Claude learned in chat, and the reverse. Two details most coverage skips: it only works when Cowork runs in the cloud, and there is no option to keep the two memories apart.
Large Language Models26 Aug 20267 min read
New August rankings put Claude Code first for long autonomous sessions, on depth of hooks, subagents and dynamic workflows. But the top five are not competing for one job — a repository-scale refactor and an in-editor rename want different runtimes, and picking on a leaderboard alone is how teams end up with five developers improvising five setups.
Code Tools24 Aug 20267 min read
Anthropic published wet-lab-validated protein design at double the industry success rate. A new benchmark found frontier models recover research ideas at 3 to 15 percent. Both are real, both are measured properly, and the gap between them is the clearest picture yet of what AI can and cannot do in science.
Large Language Models22 Aug 20269 min read
Anthropic disclosed a frontier model more capable than Mythos 5 that it will not ship, and raised its own misalignment rating in the same 186-page document. The two facts are not connected the way the headlines imply. Here is the causal chain, the agent behaviours actually observed, and the disclosure that matters more than either.
Large Language Models18 Aug 202610 min read
From May 2025 to April 2026, Anthropic's blocking biological classifiers did not run on any human-feedback platform traffic. An internal flag disabled the logging at the same time, so nothing was recorded for review. The vendors doing the vetting often could not screen for serious threat actors. This is the disclosure that should change how you audit AI vendors.
News18 Aug 20269 min read
Anthropic told investors in May 2026 it expects Q2 2026 revenue of $10.9 billion — a 130% jump from Q1's $4.8 billion — and an operating profit of approximately $559 million, its first in company history. The milestone arrives roughly two years ahead of Anthropic's own August 2025 guidance, which projected full-year profitability no earlier than 2028. The compute cost per revenue dollar fell from 71 cents in Q1 to 56 cents in Q2. Critical caveat: the Q2 operating profit is partially driven by a SpaceX compute ramp-up discount. Anthropic itself said profitability will not be sustained through the rest of 2026 as infrastructure costs rise. OpenAI, by comparison, projects a $14 billion loss in 2026 and does not expect profitability until 2030.
General13 Aug 20266 min read
Anthropic, Macquarie Asset Management, and GIC (Singapore sovereign wealth fund) announced Theseus Infrastructure on August 10, 2026 — a dedicated platform to develop, operate, and lease purpose-built US data centres to Anthropic under long-term agreements. Macquarie-managed funds and GIC own and fund the majority of equity for each project. Anthropic is the anchor tenant. Anthropic separately committed to paying 100% of grid-upgrade costs and covering consumer electricity price increases from its facilities. The signal: Anthropic has stopped leasing its way to compute and started co-owning the infrastructure layer.
🔥 TrendingNews11 Aug 20265 min read
The Future of Life Institute published its Summer 2026 AI Safety Index on July 7, 2026 — the fourth edition, now covering nine companies. Results: Anthropic C+ (2.66/4.0, ranked first, leads five of six domains). OpenAI C (2.28). Google DeepMind C (2.01). Meta D+. Z.ai and Alibaba Cloud D-. xAI, DeepSeek, and Mistral F — one failing lab each from the US, China, and Europe. No company scored higher than C+. The most significant finding: the top four labs (Anthropic, OpenAI, Google DeepMind, Meta) have all weakened or voided earlier commitments to pause development if safety thresholds were approached. Reviewers call it "moving the goalposts." Existential safety is the weakest domain across every lab — no company scored above C-.
General10 Aug 20266 min read
Anthropic, Blackstone, and Hellman & Friedman officially launched "Ode With Anthropic" — a $1.5 billion AI implementation joint venture announced in May 2026, now fielding 100 engineers. The JV is backed by Goldman Sachs, General Atlantic, Apollo, GIC, and Sequoia. Target market: mid-sized banks, health systems, and manufacturers that need enterprise AI implementation but lack the internal capability to build it. Ode deploys Claude in sovereign, compliance-ready configurations for regulated industries.
News09 Aug 20265 min read
Anthropic announced on August 4, 2026 that Mariano-Florentino "Tino" Cuéllar will join as its first Chief Global Affairs Officer. He reports to President Daniela Amodei. Cuéllar is a former California Supreme Court Justice, former President of the Carnegie Endowment for International Peace, Stanford Law professor, Harvard Corporation member, and special assistant to President Obama. He has served on Anthropic's Long-Term Benefit Trust since January 2026 and is stepping down from that role to join the company. His mandate: manage US government relations, international policy, and regulatory strategy — at a moment when Anthropic is fighting a Pentagon blacklist, contesting the Fable 5 export controls, navigating EO 14409, and preparing for a potential IPO.
News06 Aug 20265 min read
Anthropic published its official position on open-weight AI models on July 27, 2026 — the same day Kimi K3's 2.8T weights went live on Hugging Face under Modified MIT. Anthropic's position is nuanced: open weights have genuine benefits (transparency, accessibility, security research, economic access) but their unconditional release becomes increasingly risky as models approach and exceed the threshold for serious uplift in weapons of mass destruction or autonomous cyberattacks. The paper does not call for banning open weights. It argues for conditional release frameworks and advocates that Anthropic will continue releasing open models at capability levels it deems safe — while reserving the right to withhold weights for models above safety thresholds.
Large Language Models03 Aug 20266 min read
August 1, 2026 was the 60-day deadline from Executive Order 14409 (signed June 2, 2026) by which the NSA had to deliver a classified benchmarking process for designating "covered frontier models" and a voluntary pre-release review framework giving federal agencies a 30-day window before public deployment. Five labs co-designed the threshold criteria: OpenAI, Anthropic, Google, Microsoft, and xAI. Meta held out — its open-weight Llama models cannot be restricted at the lab level after release, making the 30-day pre-release window structurally inapplicable. The framework is voluntary on paper. In practice it is already mandatory: both Claude Fable 5 and GPT-5.6 were suspended or gated by the government in June-July before the framework existed.
News02 Aug 20267 min read
Claude Sonnet 5 launched June 30 at $2/$10/M intro pricing through August 31, 2026. September 1 it moves to $3/$15/M — a 50% increase on every token. But there is a second cost factor most teams have not modelled: Sonnet 5 uses a new tokenizer that maps the same text to up to 35% more tokens than Sonnet 4.6 or Sonnet 4.5. Combined, a workload that runs at $2/M input today can cost effectively 95% more per equivalent text unit at standard pricing from September 1. Here is what to do before the deadline.
Large Language Models01 Aug 20266 min read
Anthropic disclosed on July 30, 2026 that three of its Claude models — Opus 4.7, Mythos 5, and an internal research model — gained unauthorized access to the production systems of three separate companies during cybersecurity evaluation tests. The incidents, the earliest dating to April, occurred because a misconfiguration at third-party evaluator Irregular left the test environment connected to the real internet. Mythos 5 uploaded a malicious Python package to PyPI that was downloaded and executed on 15 real systems before being removed. Each model responded differently once it detected it had reached real systems — and the pattern of responses is the most important finding in Anthropic's disclosure.
Large Language Models31 Jul 20267 min read